Technical Paper

Intl J. of Aeronautical & Space Sci. 12(4), 360364 (2011) DOI:10.5139/IJASS.2011.12.4.360

An Adaptive Control Approach for Improving Control Systems with Unknown Backlash
Kwang Ho Han*, Gi Ok Koh*, Jaemin Sung** and Byoung Soo Kim**
*Flight Control Team, Research & Development Division, Korea Aerospace Industries, Ltd., Sacheon 664-710, Korea **Department of Mechanical and Aerospace Engineering, Gyeongsang National University, Jinju 660-701, Korea

Abstract
Backlash is common in mechanical and hydraulic systems and severely limits overall system performance. In this paper, the development of an adaptive control scheme for systems with unknown backlash is presented. An adaptive backlash inverse based controller is applied to a plant that has an unknown backlash in its input. The harmful effects of backlash are presented. Compensation for backlash by adding a discrete adaptive backlash inverse structure and the gradient-type adaptive algorithm, which provides the estimated backlash parameters, are also presented. The supposed adaptive backlash control algorithms are applied to an aircraft with unknown backlash in the actuator of control surfaces. Simulation results show that the proposed compensation scheme improves the tracking performance of systems with backlash. Key words:  Adaptive control, Actuator nonlinearities, Backlash, Backlash inverse

1. Introduction
Actuator devices contain nonlinearities such as friction, dead-zone, saturation, backlash, and hysteresis. Many of these nonlinearities are not continuous, but rather discontinuous functions or even dynamic models. Backlash is a dynamic nonlinearity and is common in mechanical and hydraulic systems. The undesirable effects of backlash are the main factors that severely limit the performance of feedback systems. These undesirable effects consist of non-differentiable nonlinearities and include the decrease of static output accuracy, poor transient performance, limit cycles, and instability (Santos and Vieira, 2008; Slotine and Li, 1991; Tao and Kokotovic, 1993; Tao and Kokotovic, 1996). Mechanical solutions such as spring loaded split gear assemblies and dual motor systems can satisfactorily handle the backlash problem. However, they are expensive, energy consuming, and increase the weight of the system. Therefore,
Copyright 2011. The Korean Society for Aeronautical and Space Sciences This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/bync/3.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

it is desirable to find ways to achieve backlash compensation without such mechanical devices. A commonly used approach to cancel the harmful effects of nonlinearities is the implementation of their inverse characteristics into the controller structure. A compensated inverse dynamics approach using adaptive and robust control techniques is presented in Song et al. (1994). A backlash compensation system using dynamic inversion is described in Selmic and Lewis (2001). A backlash inverse is used to reduce the harmful effects of the backlash in this paper. The parameter values of the backlash inverse are crucial to the control performance and, as such, they need to be estimated if the backlash is unknown or varies with time. This paper is organized as follows: Section 2 presents the backlash compensation, that is, the backlash model and its inverse as well as the adaptive backlash inverse. Section 3 introduces the adaptive backlash inverse control model, the controller structure, and the applied adaptive law. Section 4
** PhD Candidate ** Professor, Corresponding author, E-mail : bskim@gnu.ac.kr

Received: October 01, 2011 Accepted: December 06, 2011

Copyright The Korean Society for Aeronautical & Space Sciences

360

http://ijass.or.kr pISSN: 2093-6742 eISSN: 2093-2480

u (t ) = B(v(t )) = m(v(t ) cr ) if v(t ) vr u (t 1) if vl < v(t ) < vr

(3)

where the values vl and vr are the v-axis projections of the intersections of the two parallel lines of slope m with the horizontal inner segment containing u (t 1) (Tao and Kokotovic, 1996). Kwang Ho Han An Adaptive Control Approach for Improving Control Systems with Unknown Backlash

Fig. 4. Backlash invers

The discrete-ti represented by the

ach for Improving

analyzes the performance of our adaptive control approach through a numerical example and Section 5 presents the Control Systems with Unknown Backlash conclusions of this work.

The discrete-time version of the backlash model is also easy to visualize, as shown below: (2)

v(t ) = BI (ud (t ))

aemin Sung**, and Byoung Soo Kim**

2. Backlash Compensation
(3)

2.1 Backlash modelUniversity, Jinju 660-701, Aerospace Engineering, Gyeongsang National

elopment Division, Korea Aerospace Industries, Ltd., Sacheon 664-710, Korea

In contrast to the memoryless dead-zone, backlash has an element of memory and is dynamic. A widely accepted Abstract characteristic of backlash is shown Fig. 1, where v(t) is the and hydraulic systems and severely performance. this paper, input, u(t)limits is the overall output,system and cr>0 is the right In crossing, while ol scheme for systems with unknown backlash is presented. An adaptive backlash cl>0 is the left crossing (Tao and Kokotovic, 1996). Typically, plant that has an unknown backlash in its input. The harmful effects of backlash are the concept of backlash associated with trains as the h by adding a discrete adaptive backlash inverse is structure and the gear gradient-type schematic representation of backlash, as shown in Fig. 1b. e estimated backlash parameters, are also presented. The supposed adaptive backlash rcraft with unknown backlash in the actuator of control surfaces. Simulation The upward side is active when both v(t) results and u(t) cheme improves the tracking performance of systems with backlash. increase: (t)>0, u (t)>0 u(t) = m(v(t)-cr), v nonlinearities, Backlash, Backlash inverse The downward side is active when both v(t) and u(t) decrease: 2.uBacklash Compensation (t)<0, u (t)<0 s such as friction, (t) = m(v(t)cr), v (t ) if v (t ) > 0 and u (t ) = m(v(t ) cr ) , or mv esis. Many of these 2.1 Backlash model where m>0, c l<Cr are constant parameters. The motion on ther discontinuous In contrast memoryless dead-zone, backlash has an ( (t ) < 0 and (1) u t ) = to if v u (by t ) =u (v= ( t )0. cl )compact A m the any inner segment is characterized (t) lash is a dynamic element of memory and is dynamic. A widely accepted 0 otherwise backlash description ofof the continuous-time version of the ical and hydraulic characteristic is shown Fig. 1, where v(backlash t ) is the lash are the main B () is given in Eq.(1) u (tdiscrete-time ) is the output, and crof >0 the right crossing, input, The version the is backlash model is also ance of feedback easy to shown below:(Tao and Kokotovic, 1996). cl visualize, > 0 is theas left crossing while Typically, the concept of with gear (1) non-differentiable u (t 1) backlash is u (tassociated 1) (2) vl = representation + cl , vr = of backlash, + cr as shown in trains as the schematic of static output m m Fig. 1b. limit cycles, and e and Li, 1991; Tao m(v(t ) cl ) if v(t ) vl 996). (3) u (t ) = B(v(t )) = m(v(t ) cr ) if v(t ) vr loaded split gear u (t 1) if v < v ( t ) < v atisfactorily handle l r expensive, energy where the values vl and vr are the v-axis projections of system. Therefore, klash compensation the intersections of the two parallel lines of slope m with the horizontal inner segment containing u (t 1) (Tao and e harmful effects of Kokotovic, 1996). Fig. 1.(a) (a)Backlash Backlashmodel, model, (b) Schematic representation. of their inverse Fig. 1. Schematic representation. (t ) if v ((b) t) > 0 and u (t ) = m(v(t ) cr ) , or mv e. A compensated (1) u ( t ) if v ( t ) 0 and u ( t ) (v(t ) vc = < = mboth The upward side is active when (lt)) and u (t ) and robust control 1994). A backlash 0 otherwise increase: sion is described in (t ) > (t ) > u (t ) = mversion (v(t ) - cr of ), v 0, u 0 The discrete-time the backlash model is also easy to visualize, as shown below: e harmful effects of The downward side is active when both v(t ) and u (t ) eter values of the decrease: v = u (t 1) + c , v = u (t 1) + c erformance and, as (2) l l r r m m lash is unknown or (t ) < 0, u (t ) < 0 u (t ) = m(v(t ) - cr ), v m(v(t ) cl ) if v(t ) vl tion 2 presents the where m > 0, c < Cr are constant parameters. The ash model and its (3) u (t ) = B(v(t )) = lm(v(t ) cr ) if v(t ) vr (t ) = 0 . A motion on any innersegment is characterized by u inverse. Section 3 u ( t 1) if v < v ( t ) < v l r compact description of the continuous-time version of the control model, the ive law. Section 4 backlash ( ) B i is given in Eq.(1) v v and are the v-axis projections of where the values Fig. 2. Backlash response input. l to a saw-tooth r e control approach Fig. 2. Backlash response to a saw-tooth input. the intersections of the two parallel lines of slope m with the on 5 presents the A further insight into the nature of backlash be and gained (Tao horizontal inner segment containing u (t 1) can from the waveform of the output u (t ) when the input v(t ) Kokotovic, 1996). is the saw-tooth signal in Fig. 2.

where the values vl and vr are the v-axis projections of Fig. 2. Backlash response to a saw-tooth input. lines of slope m with the intersections of the two parallel the horizontal inner segment containing u(t-1) (Tao and A further insight into the nature of backlash can be gained Kokotovic, 1996). from the waveform of the output u (t ) when the input v(t ) A further insight into nature of backlash can be gained is the saw-tooth signal inthe Fig. 2. from the waveform of the output u(t) when the input v(t) is 2.2 saw-tooth Backlash inverse the signal model in Fig. 2. The desired function of a backlash inverse is to cancel the harmful effects of backlash on system performance. The ideal 2.2 Backlash model BI (i) will make the traverse of this segment backlash inverseinverse The desired function of athis backlash inverse is to cancel instantaneous and thus cancel undesirable backlash effect, as shown in eff Fig. 3. of That is, given desiredperformance. signal ud (t ) Th for the harmful ects backlash onasystem e u (t ) backlash , a backlash inverse BI (i)the is such of that ideal inverse BI () will make traverse this ud (t ) = B (instantaneous BI (ud (t ))) as shown in Fig. 4. segment and thus cancel this undesirable

In this paper, th inverse model is n version has the backlash inverse d (t )) for im sign(u adaptive inverse c time adaptive co knowledge is ofte modern control sys digital controllers s actual practice (Tao

2.3 Adaptive Back The backlash i approximated by upward and dow bounded gains. W

unknown, we can

backlash inverse es

(u v(t ) = BI d

In the next sec BI (i) as part of th plants with an unkn

Fig. 4. Backlash inverse response to a saw-tooth input. Fig. 3. Backlash inverse model. Fig. 3. Backlash inverse model.

3. Adaptive Bac
The discrete-time model of the backlash inverse is represented by the following mapping:
v(t 1) if ud (t ) = ud (t 1) u (t ) v(t ) = BI (ud (t )) = d + cl if ud (t ) < ud (t 1) m ud (t ) + cr if ud (t ) > ud (t 1) m

3.1 Discrete-time a In this section structure and the a of this section is t

(4)

In this paper, the continuous-time version of the backlash inverse model is not shown. This is because the discrete-time version has the following advantages. The discrete-time backlash inverse does not require knowledge of d (t )) for implementation. This makes a discrete-time sign(u adaptive inverse controller more practical than a continuoustime adaptive controller, because such signal derivative Fig. 4. Backlash inverse response to a saw-tooth input. is often unavailable in applications. In addition, Fig. knowledge 4. Backlash inverse response to a saw-tooth input. modern control systems are most frequently implemented with The discrete-time of the backlash inverse is to digital controllers somodel that a discrete-time treatment is closer represented by the (Tao following mapping: 1996). actual practice and Kokotovic,
inverse 2.3 Adaptive Backlash v(t 1) if ud (t ) = ud (t 1) The backlash inverse BI (i) defined by Eq. (4) can be ud (t ) approximated by replacing the + cl if 1) between (4) its v(t ) = BI (ud (t )) = ud vertical (t ) < ud (tjumps m upward and downward lines by continuous curves with ud (t ) bounded gains. When the backlash parameters m, cl , cr are + cr if ud (t ) > ud (t 1) m , c , c to design a unknown, we can use their estimates m r l
http://ijass.org

361

.com c.kr

2.2 Backlash inverse model The desired function of a backlash inverse is to cancel the harmful effects of backlash on system performance. The ideal backlash inverse BI (i) will make the traverse of this segment instantaneous and thus cancel this undesirable backlash effect, as shown in Fig. 3. That is, given a desired signal ud (t ) for

inverse controller to achieve asymptotic tracking, despite the presence of backlash. represented by the following mapping: presence of backlash. Let us consider a plant whose linear part is G ( s ) = k p / s , Let us consider a plant whose linear part is G ( s ) = k p / s , where k p is a known constant, assuming that the backlash isifauknown constant, assuming that the backlash where v(k tp 1) d (t ) = ud (t 1) m > 0 is known, but its width is unknown. The linear slope slope (t )> 0 is known, but its width is unknown. The linear ud m ash model ofthe plant by: (4) = v(t ) = BI (udpart (t )) clin if discrete-time uddiscrete-time (t ) < ud (t is 1)given lash model is is also also part ofm the + plant in is given by: y (t + 1) = y (t ) + u (t ) (6) ud (t ) 1) = y (t ) + u (t ) (6) Intl J. of Aeronautical & Space Sci. 12(4), 360364 (2011) + cr if ud (t ) y >(t u+ d (t 1) )+ c (2) m (2) + crr In the absence of backlash, our design objective to stabilize In the absence of backlash, our design objective to stabilize the closed loop system version and make plant output y (t ) track a In this paper, the continuous-time of the the backlash the closed loop system of and make the plant output y (t ) track a despite the presence backlash. backlash eff ect, as shown in Fig. 3. Th at is, given a desired v y ( t ) l given reference signal may be achieved by the inverse model is not shown. This is because the discrete-time vl m y (t ) may be given signal achieved the m Let reference us consider a plant whose linear part is G(s)by = kp signal ud(t) for u(t), a backlash inverse BI() is such that uhas / s, d(t) the version following advantages. The discrete-time (3) v controller: (3) vrr controller: backlash inverse does not require knowledge of = B ( BI ( u where k ( t ))) as shown in Fig. 4. is a known constant, assuming that the backlash d p v v((tt)) < <v vrr t) = but y (t ) + ym (t + 1) is unknown. The (7) d (t ))is (u for implementation. This makes a d (u The discrete-time model of the backlash sign inverse slope m>0 is u known, its width linear y( t ) discrete-time + ym (t + 1) (7) d (t ) = adaptive inverse controller more practical than a continuousaxis projections of Fig. 4. Backlash inverse response to a saw-tooth input. represented by the following mapping: part of the plant in discrete-time is given by: axis projections of Fig. 4. Backlash inverse response to a saw-tooth input. In the presence of backlash we use this controller along In the because presence such of backlash use this controller along time adaptive with controller, signal we derivative slope m the an adaptive scheme designed to update the backlash Backlash slope Fig. m 2.with with the response to a saw-tooth input. with an adaptive scheme designed to update the backlash knowledge is often unavailable in applications. In addition, (6) The discrete-time model of the backlash inverse is u The discrete-time model of the backlash inverse is inverse v(t ) = BI (u (t )) on-line, as shown in Fig. 5. (Tao and and u((tt 1) 1) (Tao modern control systems are most frequently implemented with in Fig. 5. d represented by the following mapping: v ( t ) = BI ( u ( t )) on-line, as shown inverse d represented by of thebacklash following mapping: A further insight into the nature can be gained digital controllers so that a discrete-time treatment is closer In the absence of backlash, our to design objective to from the waveform of the output u (t ) when the input v(t ) actual practice and Kokotovic, 1996). (4) (Taostabilize the closed loop system and make the plant output is the saw-tooth signal in Fig. 2. ((tt 1) if u tt)) = u tt 1) d( d( v v 1) if u ( = u ( 1) y ( t ) track a given reference signal ym(t) may be achieved by d d 2.3 Adaptive Backlash inverse u (t ) d (t ) 2.2 Backlash inverse model u the controller: d = = + < (4) v ( t ) BI ( u ( t )) c if u ( t ) u ( t 1) inverse BI (i) defined by Eq. (4) can be l if u d (t ) < u d (t 1) The backlash = + ccancel (4) v(ta )= BI (udd (t )) mis to l d the d The desired function of backlash inverse approximated by replacing the vertical jumps between its m harmful effects of backlash on system performance. ideal In this paper, the continuous-time version of the backlash (7) u udd((tt)) + c The upward and downward lines by continuous curves with u (t )) > r if >u 1) c ud udd((tt 1) discretemake the of + this segment backlash inverse BI (i) will model r if d (tbecause m inverse is traverse not shown. Th is is the bounded gains. When the backlash parameters m, cl , cr are m instantaneous and thus cancel thishas undesirable backlash effect, The discrete-time In the presence of backlash we use this controller along time version the following advantages. , c , c to design a In this paper, the the m we Fig. can 5. use their estimates In this paper, the continuous-time continuous-time version of of unknown, the backlash backlash for as shown in Fig. 3. That is, given a desired signal ud (t ) version r designed l scheme. to update the backlash with an adaptive scheme Adaptive backlash compensation backlash inverse does not require knowledge of sign(u d(t)) inverse model is not shown. This is because the discrete-time Fig. 5. Adaptive backlash compensation scheme. inverse model isBI not shown. This is because the discrete-time (u (t )) ,on-line, u (t ) , a backlash inverse ( i ) is such that inverse as shown v ( t ) = BI backlash inverse estimate: characterized by in Fig. 5. version has advantages. The d for implementation. This makes a discrete-time adaptive version has the the following following advantages. The discrete-time discrete-time m is known and cr c = cl c= = cc , , we let Since, by assumption, ud (t ) = B ( BI (ud (t )))inverse as shown in Fig. 4. backlash inverse does not of Since, by by assumption, m is is known and Since, assumption, m known and c =lcl = c we , we rr = backlashcontroller inverse more does practical not require require knowledge of than knowledge a continuous sign ( u ( t )) for implementation. This makes a discrete-time ) = m and mc ( t ) = mc ( t ) = mc ( t ) and introduce: d (t )) for implementation. This makes a discrete-time let m(t and and introduce: sign ( u l r time adaptive controller, because such signal derivative d m and mc (t ) = mcr (t ) = mc(t ) and introduce: let m(t ) = v(t 1) if ud l(t ) = ud (t 1) adaptive controller more practical than continuousadaptive inverse inverse controller more practical than a aIn continuousknowledge is often unavailable in applications. addition, t ), * = *mc time d((tt )) = (t ) * , (8) (8) * (t ) = mc ( (t ), = mc u time adaptive adaptive controller, controller, because because such such signal signal derivative derivative t )c = if (u t) , (t ) = mc (8) modern control systems are most frequently implemented (5) v(t ) = BI (ud (t )) = (+ knowledge l d (t ) < ud (t 1) knowledge is is often often unavailable unavailable in in applications. applications. In In addition, addition, m the backlash inverse error equation becomes: modern control systems are most frequently implemented with As such, with digital controllers so that a discrete-time treatment is modern control systems are most frequently implemented with As such, such, backlash inverse error equation becomes: As backlash inverse error equation becomes: lash can ud (tthe )the klash can be be gained gained digital controllers so that a discrete-time treatment is closer to c if u ( t ) u ( t 1) + > closer to actual practice (Tao and Kokotovic, 1996). digital controllers so that a discrete-time treatment is closer to r d d hen ) ud (t ) = (t ) (t ) + d b (t ) (9) u (t u hen the the input input v v((tt)) actual m (t ) ud (t ) = (t ) (t ) + d b (t ) (9) actual practice practice (Tao (Tao and and Kokotovic, Kokotovic, 1996). 1996). (9) (adaptive (t ) backlash In the next section, inverse 2.3 Adaptive Backlash inverse (use t ) = an t) wherewe regressor. 2.3 r l is the 2.3 Adaptive Adaptive Backlash Backlash inverse inverse where isthe theregressor. regressor. where (t ) = r (t ) l (t ) is ( ) BI i as part of the proposed adaptive control algorithm The backlash inverse defined by Eq. (4) can be ((ii)) de BI inverse controller to achieve asymptotic despite the For the the tracking error, e(t )e= ( t )y ymy (tfor )(t ,tracking, Eq. Th e backlash () ned by The backlash inverse inverse BI defined by Eq. Eq. (4) (4) can can be be BI For tracking error, (tt) )y= = (tt) ) ),from from Eq.(6) (6)(8), For the tracking error, e ( y ( y rse is to cancel the m m (t ) , from Eq. (6) plants with an unknown backlash . B i ( ) erse is to cancel the presence of backlash. approximated by replacing the vertical jumps between its (8), we obtain the expression: approximated by replacing the vertical jumps between its approximated by replacing the vertical jumps between its ormance. we we obtain expression: (8), obtain the expression: ormance. The The ideal ideal Let us the consider a plant whose linear part is G ( s ) = k p / s , upward and downward lines by continuous curves with upward and and downward downward lines lines by by continuous continuous 3. curves with Backlash upward curves with rse of this segment Adaptive Inverse Control erse of this segment e ( t ) = ( t 1) ( t 1) * (*t assuming 1) + db (t that 1) (10) bounded gains. When the backlash parameters m , c , c are l r bounded gains. When the backlash parameters m, cl , cr are k(pt ) = isa where e (t known 1) (t constant, 1) (t 1) + db (t 1)the backlash (10) (10) ble backlash effect, bounded gains. When the backlash parameters m , c , c are 3.1 Discrete-time adaptive Backlash inverse l r ble backlash effect, m , c , c to design a unknown, we can use their estimates An important for but use in the discrete-time adaptive d t ) for inverseunknown, m > 0 quantity is quantity known, its width is unknown. The linear slope this a section, the adaptive backlash inverse control toIn design a unknown, we we can can use use their their estimates estimates m, crr , cll to d( design Fig.u 3. Backlash model. d signal signal u An important for in the discrete-time adaptive d (t ) for An important quantity foruse use in thegoal discrete-time adaptive law is the estimation error, defined as: part of the plant in discrete-time is given by: structure by and the applied adaptive law are presented. The is such that law is the estimation error, defined as: backlash inverse estimate: , characterized by v ( t ) = BI ( u ( t )) backlash inverse estimate: characterized is such that backlash inverse estimate: v(t ) = BI (udd (t )) , characterized by is to law is the error, defined as: of this section design aestimation discrete-time adaptive backlash ) = e(t ) + (t ) (( tt 1) (t t 1) (t (11) (6) )(1) ((tt ) =y + u1) (t ) = e(t ) + (y t) (t1) t (t 1) + 1) ( (11) (11) 2 based on the Using Eq. (11), our update law for ( t ) In the absence of backlash, our design objective to stabilize Using Eq. (11), our update law for (t ) based on the ((tt 1) if u tt)) = u tt 1) v d( d( = v 1) if u ( u ( 1) d d the closedalgorithm loop system gradient-type is: and make the plant output y (t ) track a u (t ) Using Eq. (11), our gradient-type algorithm is: update law for (t) based on the (5) d (t ) u (5) ym (t ) may be achieved by the given reference signal d v ( t ) BI ( u ( t )) c if u ( t ) u ( t 1) = = + < (5) v(t ) = BI (udd (t )) = + cll if udd (t ) < udd (t 1) m ( t is: 1) (t ) t ) gradient-type algorithm ( t 1) ( ( t + 1) = ( t ) + f (t ), 0 < < 2 (12) m controller: 2 + f (t ), 0 < < 2 (12) (t + 1) = (t )1 + ( t 1) 2 u ( t ) udd (t ) + c 1 + ( t 1) if u ((tt)) > u ((tt 1) r d d c if u u 1) + > ud (t ) = y (t ) + ym (t + 1) (7) m r d d (12) m where is the constant gain and the modifying term where is the constant gain and the modifying term In the next section, we use an adaptive backlash inverse f (t ) is In the presence of backlash we use this controller along found by: In In the the next next section, section, we we use use an an adaptive adaptive backlash backlash inverse inverse f (twhere ) is found is by: the constant and the modifying f(t) BI () as part of the proposed adaptive control algorithm for with an adaptive scheme gain designed to update the term backlash ( ) BI i as part of the proposed adaptive control algorithm for BI (i) as part of the proposed adaptive control algorithm for a b is found by: 0 ( ) ( ) [ , ] if t g t + plants with an unknown backlash B (). a b ) = BI (ud (t )) on-line, as Fig. 5. ] inverse v(t plants (t shown ) + g (t )in if , [ plants with with an an unknown unknown backlash backlash B B((ii)) .. 0 b (t ) g (t ) if (t ) + g (t ) > b b f (t ) = b (13) f (t ) = (t ) g (t ) if (t ) + g (t ) > (13) 3. a a a 3. Adaptive Adaptive Backlash Backlash Inverse Inverse Control Control ( ) ( ) ( ) ( ) t g t if t g t + < a (t ) g (t ) if (t ) + g (t ) < 3.1 Discrete-time adaptive inverse 3. Adaptive Backlash Inverse Control 3.1 Discrete-time adaptive Backlash Backlash inverse In this section, the adaptive backlash inverse control In this section, the adaptive backlash inverse control structure structure and and the the applied applied adaptive adaptive law law are are presented. presented. The The goal goal 3.1 Discrete-time adaptive Backlash inverse of this section is to design a discrete-time adaptive backlash of this section is to design a discrete-time adaptive backlash In this section, the adaptive backlash inverse control 3 3 structure and the applied adaptive law are presented. The goal of this section is to design a discrete-time adaptive Fig. 5. Adaptive backlash compensation scheme. backlash inverse controller to achieve asymptotic tracking, Fig. 5. Adaptive backlash compensation scheme. Since, by assumption, m is known and cr = cl = c , we

horizontal inner segment containing ((v c v((tt)) crr)),,or or Kokotovic, 1996). (1) v (1) v((tt)) cll)) c

u (t 1)

(Tao and

g (t g

with the constants with the consta bounds of the bounds of t * = *mc : a amc b = mc : mc knowledge of mc . A knowledge of mc . mc 0 . This projectio mc 0 . This projec A detailed proof of t A detailed proof and Kokotovic, 1996) and Kokotovic, 199 properties of the closed properties of the clo

4. Numerical Exam 4. Numerical Exa

It can be seen tha It can be seen oscillations (LCOs) rel oscillations (LCOs) 2008; Ross et al., 19 2008; Ross et al., performance of feed performance of fe handling qualities. The handling qualities. T to eliminate such to eliminate such nonlinearities, backlas nonlinearities, back instability (Slotine and instability (Slotine a 2008) the adaptive co 2008) the adaptive amplitude, self indu amplitude, self i nonlinearities of Russia nonlinearities of Ru This section present presen This section pres showing limit cycles showing limit cycl order to show the ef order to show the backlash algorithms. F backlash algorithms which is the relation which is the relat deflection of control su deflection of control

with the cons Fig. 6. Flight data input o Fig. 6. Flight data inp

bounds of * = mc : a mc The backlashof charac knowledge mc The backlash cha are similar to those s are similar to proje thos mc 0 . This appear only in the spec appear only in the s A detailed proo as a form of LCO. Th as a form of LCO. and Kokotovic, 19 intentionally not includ intentionally not inc properties the cl horizontal tail of actuato horizontal tail actu order to compare the f order to compare th adaptive backlash com 4. Numerical Ex adaptive backlash c response when backl back response ba It canwhen be seen unknown backlash oscillations (LCOs unknown backlas m =2008; 1, cl = 0.08, et cal. = Ross m = 1, cl = 0.08, r c performance of handling qualities. to eliminate su nonlinearities, bac instability (Slotine 2008) the adaptive amplitude, self nonlinearities of R This section pre showing limit cyc order to show th backlash algorithm which is the rela deflection of contro

(t ) = m and mc (t ) = mc (t ) = mc (t ) and introduce: let m l r

DOI:10.5139/IJASS.2011.12.4.360

362

(t ), * = mc (t ) = (t ) * , (t ) = mc

(8)

As such, the backlash inverse error equation becomes:

u (t ) ud (t ) = (t ) (t ) + d b (t )
(t ) (t ) is the regressor. where (t ) = r l

(9)

Kwang Ho Han An Adaptive Control Approach for Improving Control Systems with Unknown Backlash

(13)

g (t ) =

(t 1) (t ) 1 + 2 (t 1)

(14)

(14)

acking, the detailed despite proof of this is givenA by Tao and Kokotovic (t 1) ( t ) and Kokotovic (Tao detailed proof of this(Tao is given by Tao g (ttracking )= (14) Kokotovic, 1996), along with the stability1996), and 1+ 2 (tthe 1) stability and tracking and Kokotovic, along with rties closed system. art is of G (the s) = k p / s loop , properties of the closed loop system. with the constants a , b being the lower and upper g that the backlash umerical Example bounds of the known backlash parameter nknown. The linear * a b can be seen that some aircraft experience limit cycle are determined from a priori : . These = mc mc : 4. Numerical Example ations (LCOs) related to actuator nonlinearities (Mattos, knowledge of mc . A natural constant is that a 0 since Ross et al., 1979). This LCO of aircraft limits the (6) (t ) 0 . (t ) of ensures that mc mc 0and . This projection rmance of feedback systems It can be degrades seen that aircraft some aircraft experience limit cycle bjective to stabilize ing qualities. Therefore, it can seen as advantageous A be detailed proof of this is given by Tao and Kokotovic (Tao oscillations (LCOs) related to actuator nonlinearities output ) track oscillations. a liminatey (tsuch One of the actuator and Kokotovic, 1996), along with the stability and tracking (Mattos, 2008; Ross et al., 1979). This LCO of aircraft nearities, backlash, can also induce limit cycles and properties of the closed loop system. achieved by the bility (Slotine and Li, 1991). According to Mattos (Mattos, limits the performance of feedback systems and degrades ) the adaptive controller is to eliminate smallTherefore, it can be seen as 4.designed Numerical Example aircraft handling qualities. tude, self induced oscillations due to actuator (7) It can be seen that some such aircraft experienceOne limitof cycle to eliminate oscillations. the nearities of Russian Su-37 advantageous aircraft. oscillations (LCOs) related to actuator nonlinearities (Mattos, actuator nonlinearities, backlash, can also induce limit his section presents example of the aircraft is controller along the numerical 2008; Ross et al., 1979). This LCO of aircraft limits the ing limit and actuator nonlinearity in flight in and Li, 1991). According to pdate the cycles backlash cycles and instability (Slotine performance of feedback systems and degrades aircraft to show the effectiveness of the proposed adaptive Mattos (Mattos, 2008) the adaptive designed to in Fig. 5. handling qualities. Therefore, it can controller be seen asis advantageous ash algorithms. Figure 6 shows the I/O of the actuator, to eliminate such oscillations. One of the eliminate amplitude, self induced oscillationsactuator due to h is the relation between control small command and the nonlinearities, backlash, can alsoSu-37 induce limit cycles and actuator of Russian aircraft. ction of control surface based on thenonlinearities flight data. instability (Slotine and Li, 1991). According to Mattos (Mattos, This section presents the numerical example of the aircraft 2008) the adaptive controller is designed to eliminate small showing limit cycles and actuator nonlinearity in fl ight in amplitude, self induced oscillations due to actuator nonlinearities ofthe Russian Su-37 aircraft. order to show effectiveness of the proposed adaptive This section presents the example the aircraft backlash algorithms. Figure numerical 6 shows the I/O ofof the actuator, showing limit cycles and actuator nonlinearity in flight in which is the relation between control the order to show the effectiveness of thecommand proposed and adaptive defl ection of control surface based on the fl ight data. backlash algorithms. Figure 6 shows the I/O of the actuator, which is the relation betweenofcontrol command and the The backlash characteristics the aircraft shown in Fig. deflection of control surface based on the flight data. 6 are similar to those shown in Fig. 1. These characteristics d cr = cl = c , we

a and upper th the constants a , b being with the the lower constants , b being the lower and upper ds of the known backlash parameter bounds of the known backlash parameter * = mc: amc a b b : . These are determined from a priori mc mc . Th ese are determined from a priori knowledge of mc. A a a ledge of mc . A natural natural constantconstant is that 0 since is that >0 since mc0. This projection (t) (t ) 0 . ensures that 0 . This projection (t ) of ensures that mc

appear only in the specific aircraft and limited flight envelope as a form of LCO. The magnitudes and flight conditions are intentionally not included. We insert the backlash model in the horizontal tail actuator of the linear longitudinal model in order to compare the flight data and review the possibility of adaptive backlash compensation. Figure 7 shows the aircraft response when backlash is applied. The parameters of unknown backlash B() are selected as follows, m =1, cl = 0.08, cr = 0.08. In order to eliminate the above harmful effects of backlash, we first apply the backlash inverse to the fixed type, not the adaptive type. This is because the signal, which is needed for the error equation for adaptive control from the aft backlash, cannot be used directly in a real aircraft control system. The detailed parameters of the fixed backlash inverse are as follows, m = 1, cl = 0.04, cr = 0.04. The differences of parameters between the backlash and the fixed backlash inverse are identified as the backlash model error and the variances from aircrafts and time. Figure 8 shows the aircraft response with the above backlash and the fixed backlash inverse. To eliminate the effects of backlash model error and the variances from aircrafts and time, we apply the adaptive backlash inverse algorithm presented in Section 3. The constant gain, of adaptive law is selected as 0.56. The signal aft backlash for adaptive law is satisfied of the purpose of this study directly. Further study are ongoing to generate the error equation using the difference of aircraft responses between the reference aircraft and the real aircraft. Fig. 9 shows the aircraft response when both backlashes are applied with the adaptive backlash inverse. It can be seen from the adaptive inverse examples that the backlash inverse parameters approximately reach the same values of the backlash parameters, once the difference between them converge to zero.

5. Conclusions
The characteristic of backlash, the mathematical model, and the backlash inducing LCO mechanism are presented in this paper. Feasibility is shown by comparison with flight data. A discrete-time adaptive backlash inverse based controller is developed for an aircraft that has an unknown backlash at its input. We verify through simulations that the backlash inverse parameter values reach the backlash parameter values in a short time interval. This means that an adaptive inverse can cancel the effect of an unknown nonlinearity, and thus improve system tracking performance. As a future project, we intend to augment the adaptive state

and introduce:
*

= mc

(8)

tion becomes:

Flight data input of actuator vs. control surface deflection.

t) (9) he backlash characteristics of the aircraft shown in Fig. 6 imilar to those shown in Fig. 1. These characteristics sor. ar only in the specific aircraft and limited flight envelope t ) , from Eq. (6) magnitudes and flight conditions are form of LCO. The ionally not included. We insert the backlash model in the ontal tail actuator of the linear longitudinal model in to compare the flight data and review the possibility of d 1) (10) b (t Fig. 6. Flight data input of actuator vs. control surface deflection. ive backlash compensation. Figure 7 shows the aircraft 6. Flight data input of actuator vs. control surface deflection. crete-time nse when adaptive backlash is Fig. applied. The parameters of own backlash B(i) are selected as follows, The backlash characteristics of the aircraft shown in Fig. 6 1, (tc 1)0.08, c(11) are similar to those shown in Fig. 1. These characteristics l = r = 0.08 . appear only in the specific aircraft and limited flight envelope 363 (t ) based on the as a form of LCO. The magnitudes and flight conditions are intentionally not included. We insert the backlash model in the horizontal tail actuator of the linear longitudinal model in order to compare the flight data and review the possibility of ), 0 < < 2 (12) adaptive backlash compensation. Figure 7 shows the aircraft response when backlash is applied. The parameters of he modifying term unknown backlash B(i) are selected as follows,

http://ijass.org

Intl J. of Aeronautical & Space Sci. 12(4), 360364 (2011)

Fig. 7. Aircraft response when backlash is applied in horizontal tail actuator.

In order to eliminate the above harmful effects of backlash, we first apply the backlash inverse to the fixed type, not the adaptive type. This is because the signal, which is needed for the error equation for adaptive control from the aft backlash, cannot be used directly in a real aircraft control system. The detailed parameters of the fixed backlash inverse are as follows, m = 1, cl = 0.04, cr = 0.04 . The differences of parameters between the backlash and the fixed backlash inverse are identified as the backlash model error and the variances from aircrafts and time. Figure 8 shows the aircraft Fig. 7.7.Aircraft response when backlash isisapplied ininhorizontal tail Fig. Aircraft response when backlash applied horizontal tail response with the above and the in fixed backlash actuator. actuator. Fig. 7.  Aircraft response when backlash backlash is applied horizontal tail inverse. actuator. In order to eliminate the above harmful effects of backlash, In order to eliminate the above harmful effects of backlash, we wefirst firstapply applythe thebacklash backlashinverse inverseto tothe thefixed fixedtype, type,not notthe the adaptive adaptivetype. type.This Thisisisbecause becausethe thesignal, signal,which whichisisneeded neededfor for the theerror errorequation equationfor foradaptive adaptivecontrol controlfrom fromthe theaft aftbacklash, backlash, cannot be used directly in aa real aircraft control system. cannot be used directly in real aircraft control system. The detailed parameters of the fixed backlash inverse are as The detailed parameters of the fixed backlash inverse are as == 1,1, clcl== 0.04, crcr== 0.04 follows, m 0.04, 0.04 . . The follows, m The differences differences of of parameters parameters between between the the backlash backlash and and the the fixed fixed backlash backlash inverse inverseare areidentified identifiedas asthe thebacklash backlashmodel modelerror errorand andthe the variances variancesfrom fromaircrafts aircraftsand andtime. time.Figure Figure88shows showsthe theaircraft aircraft response response with with the the above above backlash backlash and and the the fixed fixed backlash backlash inverse. inverse.

Fig. 9. Aircraft response when both backlash are applied with adaptive backlash inverse.

5. Conclusions
The characteristic of backlash, the mathematical model, and the backlash inducing LCO mechanism are presented in this paper. Feasibility is shown by comparison with flight data. A discrete-time adaptive backlash inverse based controller is developed for an aircraft that has an unknown backlash at its input. We verify through simulations that the backlash inverse parameter values reach the backlash parameter values in a short time interval. This means that an adaptive inverse can cancel the response effect of anboth unknown nonlinearity, and thus Fig. 9.9. Aircraft when backlash are applied with adaptive Fig. Aircraft response when both backlash are applied with adaptive improve system tracking performance. backlash inverse. backlash inverse. Fig. 9.  Aircraft response when both backlash are applied with adapAs a backlash future project, we intend to augment the adaptive state tive inverse. feedback control based on neural networks to the actuator 5. Conclusions 5. Conclusions nonlinearities of an aircraft. This approach is expected to be The of backlash, the mathematical Thecharacteristic characteristic of backlash, thenonlinearities. mathematicalmodel, model, flexible with regard to various actuator and andthe thebacklash backlashinducing inducingLCO LCOmechanism mechanismare arepresented presentedin in this paper. Feasibility isis shown by comparison with flight data. this paper. Feasibility shown by comparison with flight data. Acknowledgements A adaptive backlash based isis Adiscrete-time discrete-time adaptive backlash inverse basedcontroller controller This work is supported by the inverse Korea Aerospace Industries developed for an aircraft that has an unknown backlash at its developed for an aircraft that has an unknown backlash at its (KAI), by Priority Research Centers Program through the input. We simulations that the inverse input. Weverify verifythrough through simulations that thebacklash backlash inverse National Research Foundation of Korea (NRF) funded by the parameter reach backlash parameter values in parameter values reachthe the backlash parameter values ina a References Ministry values of Education, Science and Technology (2009short time interval. This means that an adaptive inverse can short time interval. This means that an adaptive inverse can 0094016) and the the Ministry of Knowledge Economy cancel the effect of and thus cancel the effect of an an unknown unknown nonlinearity, nonlinearity, and thus (MKE), Korea, under the Information Technology Research de Mattos, B. S. (2008). Flight Control Law Design: An improve system tracking performance. improve system tracking performance. Center (ITRC) support program supervised by the National IT Industry Perspective . Brazil: Technological Institute of As aa future project, we intend to augment the adaptive state As future project, we intend to augment the adaptive state Industry Promotion Agency (NIPA)" (NIPA-2010-[C1090feedback control based on neural networks to the actuator feedback control based on neural networks to the actuator Aeronautics. 1031-0007]). We appreciate their support and help. nonlinearities nonlinearitiesof ofan anaircraft. aircraft.This Thisapproach approachisisexpected expectedto tobe be Ross, J. L., McGirr, P. G., and Waniczek, O. J. Jr. (1979). flexible with regard to various actuator nonlinearities. References flexible with regard to various actuator nonlinearities. Flying Qualities and Flight Control System Evaluation of the Acknowledgements Acknowledgements B-1 AFFTC-TR79-2 . de Strategic Mattos, Bomber. B. S. (2008). Flight Control Law Design: An This work is supported by the Korea Aerospace Industries This work is supported by the Korea Aerospace Industries Santos, T. S. and Vieira, F. H..T. (2008). An adaptive control Industry Perspective Brazil: Technological Institute (KAI), Priority Research (KAI), by by Priority Research Centers Centers Program Program through through the the of Aeronautics. approach for discrete-time systems with unknown Backlash National Foundation of funded National Research Foundation ofKorea Korea(NRF) (NRF) funded bythe the Ross, J.Research L., McGirr, P. G., and Waniczek, O. J. Jr. by (1979). at the input. Tendencias em and Matematica Aplicada e Ministry of Education, Science Technology Ministry of TEMA Education, Science Technology (2009Flying Qualities and and Flight Control (2009System 0094016) and the Ministry of Knowledge Economy 0094016) Evaluation and ,the the the Ministry of Knowledge Economy Computacional 9, 331-340. of the B-1 Strategic Bomber. AFFTC-TR(MKE), Korea, under the (MKE), Korea, under theInformation Information Technology Research Selmic, R. R.. and Lewis, F. L. (2001).Technology Neural net Research backlash 79-2 Center (ITRC) support program supervised by National IT Center (ITRC) support program supervised bythe the National IT Santos, T. S. and Vieira, F. H. T. (2008). An adaptive control compensation with Hebbian tuning using dynamic inversion. Industry Agency (NIPA)" (NIPA-2010-[C1090Industry Promotion Promotion Agency (NIPA)" (NIPA-2010-[C1090approach for discrete-time systems with unknown Automatica , 37, 1269-1277. 1031-0007]). We appreciate their support and help. 1031-0007]). We appreciate their support and help. Backlash at the input. TEMA Tendencias em Slotine, J. J. E. and Li,Aplicada W. (1991). Applied Nonlinear Control. Matematica e Computacional , 9, 331-340. References References Englewood Prentice Selmic, R. Cliffs: R. and Lewis, Hall. F. L. (2001). Neural net backlash compensation with Hebbian tuning using Control dynamic Song, Y. D., Mitchell, T. L., and Lai, H. Y. (1994). de B. Flight Control Law Design: de Mattos, Mattos, B. S. S. (2008). (2008). Flight Control Law Design: An An inversion. Automatica , 37, 1269-1277. of a class Industry of nonlinear uncertain systems via compensated Perspective . .Brazil: Technological Perspective Brazil: Technological Institute Slotine, J.Industry J. E. and Li, W. (1991). Applied NonlinearInstitute Control . of Aeronautics. of Aeronautics. inverse dynamics approach. Transactions on Automatic Englewood Cliffs: IEEE Prentice Hall. Ross, L., McGirr, Waniczek, O. Ross, J. L., McGirr,P. P.G., G.,and andLai, Waniczek, O.J.J.Jr. Jr.(1979). (1979). Song,J. Y. D., Mitchell, T. L., and H. Y. (1994). Control of a Control , 39, 1866-1871. Flying Qualities and System Flying Qualities and Flight Flight Control Control System class of nonlinear uncertain systems via Tao, G.Evaluation and Kokotovic, P. V. (1993). Adaptive control of of the B-1 Strategic Bomber. AFFTC-TREvaluation of the B-1 Strategic Bomber. AFFTC-TRcompensated inverse dynamics approach. IEEE systems with backlash. Automatica, 29, 323-335. 79-2 . 79-2 . Transactions on Automatic Control, 39, 1866-1871. Santos, T. Vieira, H. T. (2008). An control Santos, T.S. S.and and Vieira,F. F.P. H. T. (2008). Anadaptive adaptive control Tao, G. and Kokotovic, V. (1996). Adaptive Control of approach for discrete-time systems with unknown approach for discrete-time systems with unknown Systems with Actuator and Sensor Nonlinearities. New York: Backlash Backlash at at the the input. input. TEMA TEMA Tendencias Tendencias em em Wiley. Matematica Aplicada ee Computacional ,, 9, 331-340. Matematica Aplicada Computacional 9, 331-340.

To eliminate the effects of backlash model error and the variances from aircrafts and time, we apply the adaptive backlash inverse algorithm presented in Section 3. The constant gain, of adaptive law is selected as 0.56. The feedback based on neural to signal aft control backlash for adaptive law networks is satisfied ofthe theactuator purpose of this study directly. Further This study are ongoing to generate nonlinearities of an aircraft. approach is expected tothe be error equation using the difference aircraft responses flexible with regard to various actuator of nonlinearities. between the reference aircraft and the real aircraft. Fig. 9 shows the aircraft response when both backlashes are applied with the adaptive backlash inverse. Fig. 8.8. Aircraft response both backlash are applied with fixed backlash Fig. Aircraft response both backlash are applied with fixed backlash Acknowledgements It can be seen from the adaptive inverse examples that the inverse. inverse. backlash inverse parameters approximately reach the same values of the the backlash parameters, once difference To effects of backlash model error and Toeliminate eliminate the effects of backlash modelthe error andthe the This work isconverge supported by the Korea Aerospace Industries between them to zero. variances from aircrafts and time, we apply the adaptive variances from aircrafts and time, we apply the adaptive (KAI), by Priority Research Centers Program through the backlash backlash inverse inverse algorithm algorithm presented presented in in Section Section 3. 3. The The National Research ofisis Korea (NRF) funded by adaptive selected as constant ofFoundation adaptive law law selected as 0.56. 0.56. The The constant gain, gain, of the Ministry of Education, Science and Technology (2009signal signalaft aftbacklash backlashfor foradaptive adaptivelaw lawisissatisfied satisfiedof ofthe thepurpose purpose 0094016) and the Further the Ministry ofongoing Knowledge Economy of this study directly. study are to generate the of this study directly. Further study are ongoing to generate the error using the difference error equation equation using the theInformation difference of of aircraft aircraft responses responses (MKE), Korea, under Technology Research between betweenthe thereference referenceaircraft aircraftand andthe thereal realaircraft. aircraft.Fig. Fig.99 Center (ITRC) support program supervised by the National shows showsthe theaircraft aircraftresponse responsewhen whenboth bothbacklashes backlashesare areapplied applied IT Industry Promotion Agency (NIPA) (NIPA-20104 the with adaptive backlash inverse. with the adaptive backlash inverse. ItItcan the adaptive examples that the [C1090-1031-0007]). We appreciate their support and help. canbe beseen seenfrom from the adaptiveinverse inverse examples that the backlash backlash inverse inverse parameters parameters approximately approximately reach reach the the same same values values of of the the backlash backlash parameters, parameters, once once the the difference difference between them converge to zero. between them converge to zero.

Fig. 8. Aircraft response both backlash are applied with fixed backlash inverse. Fig. 8.  Aircraft response both backlash are applied with fixed backlash inverse.

DOI:10.5139/IJASS.2011.12.4.360

Selmic, Selmic,R. R.R. R.and andLewis, Lewis,F. F.L. L.(2001). (2001).Neural Neuralnet netbacklash backlash compensation compensationwith withHebbian Hebbiantuning tuningusing usingdynamic dynamic inversion. Automatica ,, 37, 1269-1277. inversion. Automatica 37, 1269-1277. Slotine, J.J. J.J. E. and Li, W. Applied Nonlinear Control .. Slotine, E. and Li, W.(1991). (1991). Applied Nonlinear Control Englewood Cliffs: Prentice Hall. Englewood Cliffs: Prentice Hall. Song, Y. D., Mitchell, T. L., and Lai, H. Y. (1994). Control of aa Song, Y. D., Mitchell, T. L., and Lai, H. Y. (1994). Control of class class of of nonlinear nonlinear uncertain uncertain systems systems via via compensated compensated inverse inverse dynamics dynamics approach. approach. IEEE IEEE 364 Transactions on Automatic Control ,, 39, 1866-1871. Transactions on Automatic Control 39, 1866-1871.

44