This paper presents an adaptive iterative learning control (AILC) scheme for a class of nonlinear systems with unknown time-varying delays and unknown input dead-zone. A novel nonlinear form of deadzone nonlinearity is presented. The assumption of identical initial condition for ILC is removed by introducing boundary layer functions. The uncertainties with time-varying delays are compensated for with assistance of appropriate Lyapunov-Krasovskii functional and Young’s inequality. The hyperbolic tangent function is employed to avoid the possible singularity problem. According to a property of hyperbolic tangent function, the system output is proved to converge to a small neighborhood of the desired trajectory by constructing Lyapunov-like composite energy function (CEF) in two cases, while maintaining all the closed-loop signals bounded. Finally, a simulation example is presented to verify the effectiveness of the proposed approach.
1. Introduction
Practically, many engineering systems carry out repetitive tasks in fixed finite space, such as manipulators [1–3]. In high precision engineering, perfect tracking for such tasks is highly desirable. Whereas existing control approaches, such as classical PID controllers, feedback linearization, and adaptive control, may guarantee closed-loop stability, they can hardly achieve perfect tracking. Fortunately, learning and repetitive control are the alternatives to address this problem, which enhance the tracking accuracy from operation to operation for systems executing repetitive tasks. By now, iterative learning control (ILC) has become one of the most important learning strategies owing to its implementation simplicity under the repeatable control environment. The basic idea of ILC is to improve the control performance of systems from trial to trial. Traditional iterative learning controllers have been developed for more than two decades for nonlinear plants [1–12]. The control input of traditional ILC is directly updated by a learning mechanism using the information of error and input in the previous iteration, and the contraction mapping theorem is often used to analyze the stability of closed-loop systems. However, the studied systems must satisfy global Lipschitz continuous condition. Thus, there are some difficulties or limitations to apply traditional ILC for certain systems. In order to circumvent this problem, some other new ILC algorithms have been widely studied. One of the most important developments is adaptive ILC (AILC) [13–15], in which the control parameters are adjusted between successive iterations, and the so-called composite energy function (CEF) [16] is usually constructed to derive the stability conclusions. In recent years, the control community has witnessed great progress in AILC of uncertain nonlinear systems [17–27].
In practical control tasks, systems with time delays are frequently encountered. The existence of time delays may make the controllers design more complicated and challenging, especially for systems with unknown time-delays. Stabilization problem of control systems with time delay has drawn much attention [28–35] due to its mathematical challenge and application demand in real-time control. In [28–31], the controller design and stability analysis for state-delayed systems were presented and [32–35] discussed the stabilization of systems with input delays. In the field of ILC, although so many results have been obtained, only few ones were available for time-delay nonlinear systems [36–41], and the majority of these works were traditional iterative learning controllers. In the framework of AILC, Chen and Zhang [40] proposed an AILC scheme for a class of scalar systems with unknown time-varying parameters and unknown time-varying delay. In [41], an adaptive learning control design was developed for a certain class of first-order nonlinearly parameterized systems with unknown periodically time-varying delay and further extended to a class of high-order systems with both time-varying and time-invariant parameters. However, they all required the identical initial conditions on the initial states and the reference trajectory for the AILC design, which is necessary for the stability and convergence analysis but can hardly be satisfied in practical systems.
In practice, nonsmooth and nonlinear characteristics such as dead-zone, hysteresis, saturation, and backlash are common in actuator and sensors. Dead-zone is one of the most important nonsmooth and non-affine-in-input nonlinearities in many industrial processes, which can severely deteriorate system performances and give rise to extra difficulties in the controller design. Therefore, the effect of dead-zone should be taken into consideration and has been drawing much interest in the control community for a long time [42–48]. To handle the problem of unknown dead-zone in control system design, an immediate method is to construct an adaptive dead-zone inverse [42]. Continuous and discrete adaptive dead-zone inverses were built for linear systems with unmeasurable dead-zone outputs [43, 44]. Based on the assumption of the consistent dead-zone slopes in the positive and negative regions, a robust adaptive control approach was given for a class of special nonlinear systems without using the dead-zone inverse [45]. In [46, 47], the dead-zone is reconstructed into the form of a linear system with a static time-varying gain and bounded disturbances by introducing characteristic function. In [48], input dead-zone is taken into account and it is proved that the simplest ILC scheme retains its ability of achieving the satisfactory performance in tracking control. To the best of our knowledge, there is little work from the viewpoint of AILC to deal with nonlinear systems with time-delay and dead-zone nonlinearity in the literature at present stage.
In this paper, we present a novel AILC scheme for a class of nonlinear time-varying systems with unknown time-varying delays and unknown input dead-zone. To the best of our knowledge, up to now, few works have been reported in the field of AILC to deal with such kinds of systems. The main design difficulty comes from how to deal with dead-zone nonlinearity and delay-dependent uncertainty. In our work, the dead-zone output is represented as a novel simple nonlinear system with a time-varying gain, which is more general than the linear form in [36]. The approach removes the assumption of linear function outside the dead-band without necessarily constructing a dead-zone inverse. An appropriate Lyapunov-Krasovskii functional and Young’s inequality are combined to eliminate the unknown time-varying delays such that the design of the control law is free from these uncertainties. Furthermore, the possible singularity which may be caused by the appearance of the reciprocal of tracking error is avoided by employing the hyperbolic tangent function. By constructing a Lyapunov-like CEF, the stability conclusion is obtained in two cases by exploiting the properties of the hyperbolic tangent function via a rigorous analysis. In addition, the boundary layer function is introduced to remove the requirement for identical initial condition which is required for the majority of ILC schemes.
The rest of this paper is organized as follows. The problem formulation and preliminaries are given in Section 2. The AILC design is developed in Section 3. The CEF-based stability analysis is presented in Section 4. A simulation example is presented to verify the validity of the proposed scheme in Section 5, followed by conclusions in Section 6.
2. Problem Formulation and Preliminaries2.1. Problem Formulation
Consider a class of nonlinear time-varying systems with unknown time-varying time-delays and dead-zone running on a finite time interval [0,T] repeatedly which is given by
(1)x˙i,k(t)=xi+1,k(t),i=1,…,n-1,x˙n,k(t)=f(Xk(t),Xτ,k(t),θ(t))+b(t)uk(t)+d(t),yk(t)=x1,k(t),uk(t)=D(vk(t)),t∈[0,T],xi,k(t)=ϖi(t),t∈[-τmax,0),i=1,…,n,
where t is the time, k∈N denotes the times of iteration, N is the integer set and denotes the sets of iteration times, yk(t)∈R and xi,k(t)∈R,i=1,…,n are the system output and states, respectively, Xk(t)≜[x1,k(t),…,xn,k(t)]T is the state vector, τ(t) is unknown time-varying delay of states and xi,kτ≜xi,k(t-τ(t)), i=2,…,n, and Xτ,k(t)=[x1,kτ(t),…,xn,kτ(t)]T,f(·,·,·) are unknown smooth functions, and b(t) is the unknown continuous time-varying gain of the system input. θ(t) is unknown continuous time-varying parameter vector; d(t) is unknown bounded external disturbance. ϖi(t) denote the initial functions for delayed states, i=1,…,n. Consider vk(t)∈R is the control input and the actuator nonlinearity D(vk(t)) is described as a dead-zone characteristic.
In this paper, a reference trajectory vector is given by Xd(t)=[yd(t),y˙d(t),…,yd(n-1)(t)]T. The tracking error vector is ek(t)=[e1,k,e2,k,…,en,k]T=Xk(t)-Xd(t). The control objective of this paper is to design an adaptive iterative learning controller uk(t), such that the tracking error ek(t) converges to a small neighborhood of the origin as k→∞; that is, limk→∞∥ek(t)∥≤εe∞, while all the signals in the closed-loop system remain bounded, where εe∞ is a small positive error tolerance which will be given in the subsequent context and ∥·∥ denotes the Euclidian norm. Define the filtered tracking error as esk(t)=[ΛT1]ek(t), where Λ=[λ1,λ2,…,λn-1]T and λ1,…,λn-1 are the coefficients of Hurwitz polynomial H(s)=sn-1+λn-1sn-2+⋯+λ1.
To facilitate control system design, we make the following reasonable assumptions for the system functions, unknown time delays, and reference signals.
Assumption 1.
The unknown state time-varying delay τ(t) satisfies 0≤τ(t)≤τmax, τ˙(t)≤κ<1, where τmax and κ are unknown positive constants.
Assumption 2.
The unknown smooth functions f(·,·,·) satisfy inequality
(2)|f(Xk,Xτ,k,θ(t))-f(Xd,Xd,τ,θ(t))|≤∥Xk-Xd∥h1(Xk,Xd)ξ1(θ)+∥Xτ,k-Xd,τ∥h2(Xτ,k,Xd,τ)ξ2(θ),
where Xd,τ≜Xd(t-τ(t)), h1(·,·) and h2(·,·) are known positive smooth functions, and ξ1(θ) and ξ2(θ) are unknown smooth functions of θ(t).
Assumption 3.
The sign of b(t) is known; without loss of generality, we always assume b(t)>0.
Assumption 4 (see [23]).
The initial state errors ei,k(0) at each iteration are not necessarily zero small and fixed, but they are assumed to be bounded.
Assumption 5.
The reference state trajectory Xd(t) is continuous, bounded, and available.
Assumption 6.
The unknown external d(t) is bounded; that is, |d(t)|≤dmax with an unknown constant dmax.
Remark 7.
Assumption 1 is common in the control problem of time-varying delay systems, which guarantees that the time delay terms can be eliminated by using Lyapunov-Krasovskii functional. Moreover, Assumption 1 is milder than that in [39–41] as it does not require the true value of τmax and κ.
Remark 8.
As b(t) is continuous on [0,T], there exist constants 0<bmin≤bmax such that bmin≤b(t)≤bmax. However, the control gain bounds bmin and bmax are only required for analytical purposes; their true values are not necessarily known in the sense that they are not used for controller design.
2.2. Dead-Zone Characteristic
The dead-zone characteristic can be described as
(3)uk(t)=D(vk(t))={m(t)(vk(t)-br)forvk(t)≥br,0forbl<vk(t)<br,m(t)(vk(t)-bl)forvk(t)≤bl,
where br≥0 and bl≤0 are unknown constants, m(t)>0 is unknown time-varying slopes, and vk(t) is the input and uk(t) is the output of dead-zone. A graphical representation of the dead-zone in this paper is shown in Figure 1.
Dead-zone model.
The dead-zone output uk(t) is not available for measurement. We make the following assumption on the dead-zone parameters.
Assumption 9.
The dead-zone parameters br, bl, and m(t) are bounded. That is, there exist unknown constants brmin, brmax, blmin, blmax, mmin, and mmax, such that brmin≤br≤brmax, blmin≤bl≤blmax, and mmin≤m(t)≤mmax.
From a practical point of view, we can redefine the dead-zone nonlinearity as
(4)uk(t)=D(vk)=m(t)vk(t)-d1(vk(t))
with
(5)d1(vk(t))={m(t)brforvk(t)≥br,m(t)vk(t)forbl<vk(t)<br,m(t)blforvk(t)≤bl.
It is obvious that d1(vk(t)) is bounded.
Remark 10.
Obviously, the dead-zone characteristic is nonlinear. And the form in [36] is the special case of (3) when m(t) is invariant. Therefore, the presentation of dead-zone in our work is more general than the earlier results.
2.3. A Motivating Example
In order to clarify the main idea of AILC, we show the design procedure briefly by a simple scalar system running on [0,T] as follows:
(6)z˙k(t)=θ(t)ξ(zk,t)+ukz(t),
where zk(t) and ukz(t) are the system state and the control input in the kth iteration, respectively, θ(t) is an unknown time-varying parameter, and ξ(zk,t) is a known time-varying function. The reference trajectory is zr(t), t∈[0,T]. Define the tracking error as ekz(t)=zk(t)-zr(t) and design the control law and adaptive learning law for the unknown time-varying parameter for the kth iteration as follows:
(7)ukz(t)=-k1ekz(t)+z˙r(t)-θ^k(t)ξ(zk,t),θ^k(t)=θ^k-1(t)+qξ(zk,t)ekz(t),θ^0(t)=0,t∈[0,T],
where k1,q>0 are design parameters. Define the estimate error as θ~k(t)=θ^k(t)-θ(t). Choose a Lyapunov-like CEF as
(8)Ekz(t)=12ekz(t)+12q∫0tθ~k(σ)dσ.
Throughout this paper, σ denotes the integral variable. Then it can be derived that
(9)ΔEkz(t)=Ekz(t)-Ek-1z(t)≤-∫0t(ekz(σ))2dσ.
We can further derive that
(10)limk→∞∫0T(ekz(σ))2dσ=0.
Therefore, the system state zk(t) converges to the reference trajectory zr(t) on [0,T] as k→∞.
3. AILC Design
According to Assumption 4, we know that there exist known constants εi such that |ei,k(0)|≤εi, i=1,2,…n for any k∈N. In order to relax the identical initial condition in ILC, we employ a boundary layer function [23] as follows:
(11)sk(t)=esk(t)-η(t)sat(esk(t)η(t)),(12)η(t)=εe-Kt,K>0,
where ε=[ΛT1][ε1,ε2,…,εn]T and K is a design parameter. The saturation function sat(·) is given by
(13)sat(esk(t)η(t))={1,ifesk(t)>η(t),esk(t)η(t),if|esk(t)|≤η(t),-1,ifesk(t)<-η(t).
Remark 11.
Note that η(t) decreases along time axis with initial condition η(0)=ε and 0<η(T)≤η(t)≤ε, ∀t∈[0,T], and then if sk(t) can be derived to zero ∀t∈[0,T], the states will asymptotically converge to the reference trajectory for all t∈[0,T].
It can be easily shown that
(14)|esk(0)|=|λ1e1,k(0)+λ2e2,k(0)+⋯+en,k(0)|≤λ1|e1,k(0)|+λ2|e2,k(0)|+⋯+|en,k(0)|≤λ1ε1+λ2ε2+⋯+εn=ε=η(0)
which implies that sk(0)=esk(0)-η(0)(esk(0)/η(0))=0 is satisfied forall k∈N. For the subsequent controller design, we firstly give the dynamic of en,k(t) as follows:
(15)e˙n,k(t)=f(Xk(t),Xτ,k,θ(t))+b(t)uk+d(t)-yd(n)(t)=f(Xk,Xτ,k,θ(t))-f(Xd,Xd,τ,θ(t))+f(Xd,Xd,τ,θ(t))+b(t)(m(t)vk(t)-d1(vk(t)))+d(t)-yd(n)(t)=f(Xk,Xτ,k,θ(t))-f(Xd,Xd,τ,θ(t))+f(Xd,Xd,τ,θ(t))+b(t)m(t)vk(t)+d2(t)-yd(n)(t),
where d2(t)=-b(t)d1(vk(t))+d(t). By Assumptions 3 and 6, we know that d2(t) is bounded; that is, there exists an unknown smooth positive function d-(t) such that |d2(t)|≤d-(t). For the simplicity of expression, we define bm(t)=b(t)m(t), Θ(t)=f(Xd,Xd,τ,θ(t)), and Δk(t)=f(Xk,Xτ,k,θ(t))-f(Xd,Xd,τ,θ(t)). It is clear that Θ(t) is an unknown time-varying function which is invariant in the iteration domain and b_m=mminbmin≤bm(t)≤mmaxbmax=b-m. Define a smooth scalar function as
(16)Vsk(t)=12sk2(t).
Differentiating Vsk(t) with respect to time, we can obtain
(17)V˙sk(t)=sk(t)s˙k(t)={sk(t)(e˙sk(t)-η˙(t)),ifesk(t)>η(t)0,if|esk(t)|≤η(t)sk(t)(e˙sk(t)+η˙(t)),ifesk(t)<-η(t)=sk(t)(e˙sk(t)-η˙(t)sgn(sk(t)))=sk(t)(∑j=1n-1λjej+1,k(t)-η˙(t)sgn(sk(t))+Θ(t)+Δk(t)+bm(t)vk(t)∑j=1n-1λjej+11,k(t)+d2(t)-yd(n)(t)∑j=1n-1λjej+1,k(t))=sk(t)(∑j=1n-1λjej+1,k(t)+Kη(t)sgn(sk(t))+Kesk(t)-Kesk(t)+Θ(t)+Δk(t)+bm(t)vk(t)+d2(t)-yd(n)(t)∑j=1n-1λjej+1,k(t))=sk(t)(Θ(t)+Δk(t)+bm(t)vk(t)+μk(t)+d2(t))-Ksk2(t),
where μk(t)=∑j=1n-1λjej+1,k(t)+Kesk(t)-yd(n)(t), and we use the relation
(18)sk(t)(-Kesk(t)+Kη(t)sgn(sk(t)))=sk(t)(-Ksk(t)-Kη(t)sat(esk(t)η(t))+Kη(t)sgn(sk(t))(esk(t)η(t)))=-Ksk2(t)-Kη(t)|sk(t)|+Kη(t)|sk(t)|=-Ksk2(t).
Utilizing Young’s inequality and noting Assumption 2, it follows that
(19)sk(t)Δk(t)≤|sk(t)|×(+∥Xτ,k-Xd,τ∥h2(Xτ,k,Xd,τ)ξ2(θ)∥Xk-Xd∥h1(Xk,Xd)ξ1(θ)+∥Xτ,k-Xd,τ∥h2(Xτ,k,Xd,τ)ξ2(θ))≤12sk2(t)ξ12(θ)+12∥ek∥2h12(Xk,Xd)+12sk2(t)ξ22(θ)+12∥eτ,k∥2h22(Xτ,k,Xd,τ).
Substituting (19) into (17) leads to
(20)V˙sk(t)≤sk(t)×(12Θ(t)+bm(t)vk(t)+μk(t)+d2(t)+12sk(t)ξ12(θ)+12sk(t)ξ22(θ))-Ksk2(t)+12∥ek∥2h12(Xk,Xd)+12∥eτ,k∥2h22(Xτ,k,Xd,τ).
To overcome the design difficulty arising from the unknown time-varying delay term, consider the following Lyapunov-Krasovskii functional:
(21)VUk(t)=12(1-κ)∫t-τ(t)t∥ek(σ)∥2h22(Xk(σ),Xd(σ))dσ.
Recalling Assumption 1, taking the time derivative of VUk(t) leads to
(22)V˙Uk(t)=12(1-κ)∥ek∥2h22(Xk,Xd)-1-τ˙(t)2(1-κ)∥eτ,k∥2h22(Xτ,k,Xd,τ)≤12(1-κ)∥ek∥2h22(Xk,Xd)-12∥eτ,k∥2h22(Xτ,k,Xd,τ).
Define a Lyapunov functional as Vk(t)=Vsk(t)+VUk(t); combining (20) and (22), we can obtain
(23)V˙k(t)≤sk(t)×(12Θ(t)+bm(t)vk(t)+μk(t)+d2(t)+12sk(t)ξ12(θ)+12sk(t)ξ22(θ))-Ksk2(t)+12∥ek∥2h12(Xk,Xd)+12(1-κ)∥ek∥2h22(Xk,Xd).
For the convenience of expression, denote ζk(t)=(1/2)∥ek∥2h12(Xk,Xd)+(1/2(1-κ))∥ek∥2h22(Xk,Xd) then (23) can be simplified as
(24)V˙k(t)≤sk(t)×(ζk(t)sk(t)Θ(t)+bm(t)vk(t)+μk(t)+d2(t)+12sk(t)ξ12(θ)+12sk(t)ξ22(θ)+ζk(t)sk(t))-Ksk2(t).
Here, we note that singularity problem may occur in (24) due to the term ζk(t)/sk(t) which approaches ∞ as sk(t) approaches zero. In order to tackle this problem, we exploit the following characteristic of hyperbolic tangent function.
Lemma 12 (see [49]).
For any constant η>0 and any variable p∈R,
(25)limp→0tanh2(p/η)p=0.
By employing the hyperbolic tangent function, (24) can be rewritten as
(26)V˙k(t)≤sk(t)×(12Θ(t)+d2(t)+bm(t)vk(t)+μk(t)+12sk(t)ξ12(θ)+12sk(t)ξ22(θ)+a2sk(t)tanh2(sk(t)η(t))∥ek∥2h12(Xk,Xd)+a2(1-κ)sk(t)tanh2(sk(t)η(t))×∥ek∥2h22(Xk,Xd)12)-Ksk2(t)+(1-atanh2(sk(t)η(t)))ζk(t),
where a>1 is a constant. From Lemma 12, we know that limsk(t)→0(a/sk(t))tanh2(sk(t)/η(t))ζk(t)=0. Hence, (a/sk(t))tanh2(sk(t)/η(t))ζk(t) is defined at sk(t)=0 and the possible singularity problem has been avoided. Upon multiplication of (26) by 1/bm(t), it becomes
(27)V˙k(t)bm(t)≤sk(t)×(1bm(t)(Θ(t)+d2(t))+vk(t)+1bm(t)μk(t)+12bm(t)sk(t)(ξ12(θ)+ξ22(θ))+a2bm(t)sk(t)tanh2(sk(t)η(t))×∥ek∥2h12(Xk,Xd)+a2bm(t)(1-κ)sk(t)tanh2(sk(t)η(t))×∥ek∥2h22(Xk,Xd)1bm(t))-Kbm(t)sk2(t)+1bm(t)×(1-atanh2(sk(t)η(t)))ζk(t)=sk(t)(βT(t)Φk(t)+vk(t))-Kbm(t)sk2(t)+1bm(t)(1-atanh2(sk(t)η(t)))ζk(t),
where β(t)=[(1/bm(t))(Θ(t)+d2(t)), 1/bm(t), (1/bm(t))(ξ12(θ)+ξ22(θ)), (1/(1-κ)bm(t))]T denote the unknown time-varying parameter vector that is invariant along the iteration axis and Φk(t)=[1, μk(t)+(a/2sk(t))tanh2(sk(t)/η(t))∥ek∥2h12(Xk,Xd),sk(t), (a/2sk(t))tanh2(sk(t)/η(t))∥ek∥2h22(Xk,Xd)]T. Based on (27), we can design the adaptive iterative learning controller as follows:
(28)vk(t)=-β^k(t)Φk(t)-K1sk(t),
where K1>0 is design parameters and β^k(t) is the estimate of β(t) in the kth iteration. The adaptive learning algorithms for unknown parameter are given by
(29)β^k(t)=β^k-1(t)+qsk(t)Φk(t),β^0(t)=0,t∈[0,T],
where q>0 is the learning gain. Define the estimation error as β~k(t)=β^k(t)-β(t). Hence, substituting the controller (28) back into (27) yields
(30)V˙k(t)bm(t)≤-sk(t)β~kT(t)Φk(t)-(Kbm(t)+K1)sk2(t)+1bm(t)(1-atanh2(sk(t)η(t)))ζk(t)≤-sk(t)β~kT(t)Φk(t)-(Kb-m+K1)sk2(t)+1bm(t)(1-atanh2(sk(t)η(t)))ζk(t).
For simplicity in expression, we denote (K/b-m+K1) by K2=(K/b-m+K1). Then, (30) can be continued as
(31)sk(t)β~kT(t)Φk(t)≤-V˙k(t)bm(t)-K2sk2(t)+1bm(t)(1-atanh2(sk(t)η(t)))ζk(t).
4. Stability and Convergence Analysis
In this section, we will check the stability of the closed-loop system and the convergence of tracking errors by CEF-based analysis. First of all, we give the following property of the tangent hyperbolic function.
Lemma 13.
Define a compact set Ωsk as Ωsk:={sk(t)∣|sk(t)|≤mη(t)}. Then, for any sk(t)∉Ωsk, the following inequality holds:
(32)1-atanh2(sk(t)η(t))<0,
where m=ln(a/(a-1)+1/(a-1)).
Proof.
See the appendix.
The stability and convergence property of the proposed AILC scheme is summarized as follows.
Theorem 14.
Considering closed-loop system (1), if Assumptions 1–6 and 9 hold, designing the control laws (28) with adaptive updating laws (29), the following properties can be guaranteed: (i) all the signals of the closed-loop system are bounded; (ii) the filtered tracking error esk(t) converges to a small neighborhood of zero as k→∞ in L2 norm; that is, limk→∞∫0T(esk(σ))2dσ≤εesk, εesk=(1/2K)(1+m)2ε2; and (iii) the tracking error vector satisfies limk→∞∥ek(t)∥≤εe∞, εe∞=(1+∥Λ∥)(k0∑i=1n-1εi+(1/(λ0-K))(1+m)εk0)+(1+m)ε, where λ0 and k0 are positive constants and will be given later.
Proof.
Define a Lyapunov-like CEF as follows:
(33)Ek(t)=12q∫0tβ~kT(σ)β~k(σ)dσ.
The difference of Ek(t) is
(34)ΔEk(t)=Ek(t)-Ek-1(t)=12q∫0t(β~kT(σ)β~k(σ)-β~k-1T(σ)β~k-1(σ))dσ.
Utilizing the algebraic relation (a-b)T(a-b)-(a-c)T(a-c)=(c-b)T[2(a-b)+(b-c)] and taking adaptive learning law (29) into account, we have the following inequality:
(35)ΔEk(t)=∫0tsk(σ)β~kT(σ)Φk(σ)dσ-q2∫0tsk2(σ)∥Φk(σ)∥2dσ≤∫0tsk(σ)β~kT(σ)Φk(σ)dσ.
Substituting (31) into (35), it follows that
(36)ΔEk(t)≤-∫0tV˙k(σ)bm(σ)dσ+∫0t1bm(σ)(1-atanh2(sk(σ)η(σ)))ζk(σ)dσ-∫0tK2sk2(σ)dσ≤-1b-mVk(t)-K2∫0tsk2(σ)dσ+∫0t1bm(σ)(1-atanh2(sk(σ)η(σ)))ζk(σ)dσ.
For analysis of stability, we consider two cases.
Case1. Consider (sk(t)∈Ωsk). When sk(t)∈Ωsk, |sk(t)|≤mη(t) holds. If sk(t)=0, we know esk(t) is bounded by η(t); that is, |esk(t)|≤η(t). If sk(t)>0, we have sk(t)=esk(t)-η(t), and from |sk(t)|≤mη(t) we can obtain sk(t)=esk(t)-η(t)≤mη(t) which further implies 0<esk≤(1+m)η(t). Similarly, if sk(t)<0, we have sk(t)=esk(t)+η(t)≥-mη(t) which means 0>esk(t)≥-(1+m)η(t). Synthesizing the above analysis, we know that |esk(t)|≤(1+m)η(t) holds. Obviously, xi,k(t) are bounded since Xd(t) is bounded. According to the smoothness of h1(·,·) and h2(·,·), we know that Φk(t) is a bounded vector. Recalling updating law (29), we know β^0(t)=0, t∈[0,T]; then, when sk(t)∈Ωsk, β^k(t) is bounded as well, k∈N. Following this chain of reasoning, the boundedness of vk(t) can be deduced. As such, all closed-loop signals are bounded.
Remark15. Theoretically, m can be made arbitrarily small by choosing a, for example, when we choose a=100 and m=0.099. This leads to sk arbitrarily small. However, large a may give rise to high gain control which can deteriorate the transient performance of closed-loop system. Consequently, in practical applications, the designers should choose appropriate design parameters to gain satisfactory transient performance and the ideal tracking error.
Case2. Consider (sk(t)∉Ωsk). According to Lemma 13, we know that the last term of ΔEk(t) can be removed from the analysis. Therefore, (36) can be simplified as
(37)ΔEk(t)≤-1b-mVk(t)-K2∫0tsk2(σ)dσ<0.
Inequality (37) shows that Ek(t) is decreasing along iteration axis. Thus, the boundedness of Ek(t) can be guaranteed as long as E1(t) is finite. According to the definition Ek, E1(t) is given by
(38)E1(t)=12q∫0tβ~1T(σ)β~1(σ)dσ.
Taking the time derivative of E1(t) results in
(39)E˙1(t)=12qβ~1T(t)β~1(t).
Recalling parameter adaptive laws (29), we have β^1(t)=qs1(t)Φ1(t), and then we obtain
(40)E˙1(t)=12qβ~1T(t)β~1(t)=12q(β~1T(t)β~1(t)-2β~1T(t)β^1(t))+1qβ~1T(t)β^1(t)=12q((β^1(t)-β(t))T(β^1(t)-β(t))-2(β^1(t)-β(t))Tβ^1(t))+s1(t)β~1T(t)Φ1(t)=12q(-β^1T(t)β^1+βT(t)β(t))+s1(t)β~1T(t)Φ1(t).
Substituting (31) into (40) yields
(41)E˙1(t)≤-V˙1(t)bm(t)-K2s12(t)+12qβT(t)β(t).
Denote βmax=maxt∈[0,T]{(1/2q)βT(t)β(t)}. Integrating the above inequality over [0,t] leads to
(42)E1(t)-E1(0)≤-1b-mV1(t)-K2∫0ts12(σ)dσ+t·βmax.
Obviously, E1(0)=0, and then it follows from (42) that
(43)E1(t)≤t·βmax<∞
which indicates the boundedness of E1(t), so Ek(t) is finite for any k∈N. Using (37) repeatedly, we have
(44)Ek(t)=E1(t)+∑l=2kΔEl(t)<E1(t)-1b-m∑l=2kVl(t)-∑l=2kK2∫0tsl2(σ)dσ≤E1(t)-∑l=2kK2∫0tsl2(σ)dσ.
We rewrite inequality (44) as
(45)∑l=2kK2∫0tsl2(σ)dσ≤(E1(t)-Ek(t))≤E1(t).
Let t=T, and, taking the limitation of (45), it follows that
(46)limk→∞∑l=2k∫0Tsl2(σ)dσ≤1K2E1(T).
Since E1(T) is bounded, with the aid of the convergence theorem of the sum of series, limk→∞∫0Tsk2(σ)dσ=0, which implies that limk→∞sk(t)=s∞(t)=0, ∀t∈[0,T]. Moreover, from definition (11), we can know that, when |esk(t)|≤η(t), sk(t)=0, then limk→∞∫0Tsk2(σ)dσ=0 is equivalent to limk→∞|esk(t)|≤η(t), which furthermore implies that limk→∞∫0T(esk(σ))2dσ≤∫0T(η(σ))2dσ.
According to the boundedness of Ek(t), we can obtain the boundedness of β^k(t). From ∫0tsk2(σ)dσ≤∫0Tsk2(σ)dσ, we can get the boundedness of sk(t). Considering the finiteness of reference trajectory Xd(t), we further obtain that xi,k(t) are bounded. Based on the above reasoning, we can obtain the boundedness of vk(t) by similar analysis in Case 1.
Synthesizing the derivations in two cases, we can conclude that the proposed control algorithm is able to guarantee that all closed-loop signals are bounded and limk→∞|esk(t)|≤(1+m)η(t). Therefore, we can obtain limk→∞∫0T(esk(σ))2dσ≤εe, εe=∫0T((1+m)η(σ))2dσ=(1/2K)(1+m)2ε2(1-e-2KT)≤1/2K)(1+m)2ε2=εesk; thus, the control objective is achieved. Furthermore, the bound of es∞(t) will satisfy limk→∞|esk(t)|=es∞(t)=(1+m)εe-Kt, ∀t∈[0,T].
Define the vector ψk(t)=[e1,k(t),e2,k(t),…,en-1,k(t)]T, and then a state representation of esk(t)=[ΛT1]ek(t) can be expressed as
(47)ψ˙k(t)=Asψk(t)+bsesk(t),
where
(48)As=[01⋯0⋮⋮⋱⋮00⋯1-λ1-λ2⋯-λn-1]∈R(n-1)×(n-1),bs=[0⋮01]∈Rn-1
with As as a stable matrix. In addition, there are two constants k0>0 and λ0>0 such that ∥eAst∥≤k0e-λ0t [50]. The solution for ψ˙k(t) is
(49)ψk(t)=eAstψk(0)+∫0teAs(t-σ)bs|esk(σ)|dσ.
Accordingly, it follows from (49) that
(50)∥ψk(t)∥=k0∥ψk(0)∥e-λ0t+k0∫0te-λ0(t-σ)|esk(σ)|dσ.
When we choose suitable parameters such that λ0>K, from limk→∞|esk(t)|≤(1+m)η(t), we can have
(51)∥ψ∞(t)∥=k0∥ψ∞(0)∥e-λ0t+k0∫0te-λ0(t-σ)|es∞(σ)|dσ≤k0∥ψ∞(0)∥+(1+m)εk0∫0te-λ0(t-σ)e-Kσdσ=k0∥ψ∞(0)∥+(1+m)εk01λ0-K(e-Kt-e-λ0t)≤k0∥ψ∞(0)∥+1λ0-K(1+m)εk0.
Noting esk(t)=[ΛT1]ek(t) and ek(t)=[ψkT(t)en,k(t)]T, we have
(52)∥ek(t)∥≤∥ψk(t)∥+|en,k(t)|=∥ψk(t)∥+|esk(t)-ΛTψk(t)|≤(1+∥Λ∥)∥ψk(t)∥+|esk(t)|.
Combining the previous two inequalities, we can obtain
(53)∥e∞(t)∥≤(1+∥Λ∥)∥ψ∞(t)∥+|es∞(t)|≤(1+∥Λ∥)(k0∥ψ∞(0)∥+1λ0-K(1+m)εk0)+(1+m)η(t)≤(1+∥Λ∥)(k0∑i=1n-1εi+1λ0-K(1+m)εk0)+(1+m)η(t)≤(1+∥Λ∥)(k0∑i=1n-1εi+1λ0-K(1+m)εk0)+(1+m)ε=εe∞.
This concludes the proof.
5. Simulation Studies
In this section, a simulation example is presented to verify the effectiveness of the AILC scheme. Consider the following second-order nonlinear system with unknown time-varying delays and unknown dead-zone running on [0,10], repetitively:
(54)x˙1,k(t)=x2,k(t),x˙2,k(t)=f(Xk(t),Xτ,k(t),θ(t))+b(t)uk(t)+d(t),yk(t)=x1,k(t),uk(t)=D(vk(t)),
where f(Xk(t),Xτ,k(t),θ(t))=-(x1,k(t)+x2,k(t))θ(t)+exp(-θ(t)((x1,kτ(t))2+(x2,kτ(t))2)), b(t)=2+0.5sint, d(t)=0.1sint, and the time delays are τ(t)=0.5(1+sint) with τmax=1, θ(t)=|cos(t)|. It can be easily verified that
(55)|exp(-θ(t)∥Xk∥2)-exp(-θ(t)∥Xd∥2)|≤∥Xk-Xd∥2|θ(t)|e-0.5.
Obviously, Assumptions 1–3 and Assumptions 5, 6, and 9 are satisfied. Moreover, we can know that h1=1 and h2=1. We give the simulation study in the following three cases.
Case 1.
The reference trajectory to be tracked by the state vector is given by Xd(t)=[sint,cost]T. The design parameters are chosen as ε1=ε2=1, λ=2, K=3, γ=0.5, K1=2, q=1, a=5, ε=λε1+ε2=3. The parameters for dead-zone are specified by m=1+0.2sint, br=0.25, bl=-0.25. The initial conditions x1,k(0) and x2,k(0) are randomly taken in the intervals [-0.5,0.5] and [0.5,1.5], respectively. Parts of the simulation results are shown in Figures 2, 3, 4, 5, and 6. From the simulation results, we can see that the proposed AILC is effective in the sense that it can drive the tracking errors converge to zero along the iteration axis.
System output y on yd(k=1) in Case 1.
System output y on yd(k=10) in Case 1.
Control input uk and vk(k=1) in Case 1.
Control input uk and vk(k=10) in Case 1.
∫0Tsk2(t)dt versus the number of iterations in Case 1.
Case 2.
To show the control performance for more complicated reference trajectory, we choose the reference trajectory as Xd(t)=[sint+sin(1.5t),cost+1.5cos(1.5t)]T. The design parameters are chosen the same as those in case 1. The initial conditions x1,k(0) and x2,k(0) are randomly taken in the intervals [-0.5,0.5] and [2,3], respectively. Parts of the simulation results are shown in Figures 7, 8, 9, 10, and 11. It shows that for more complicated reference trajectory the proposed approach is also able to achieve excellent tracking performance.
System output y on yd(k=1) in Case 2.
System output y on yd(k=10) in Case 2.
Control input uk and vk(k=1) in Case 2.
Control input uk and vk(k=10) in Case 2.
∫0Tsk2(t)dt versus the number of iterations in Case 2.
Case 3.
Finally, the contribution of this paper is shown by comparing the proposed controller with traditional adaptive controller. The controller is the same, but the adaptive laws using σ-modification for parameters are given by
(56)β^˙(t)=-Γ[Φk(t)sk+σβ^(t)],β^(0)=0.
The design parameters are given by Γ=diag{0.01} and σ=0.5. Since traditional adaptive controller does not run repeatedly, the notation k in this case does not have any practical meaning. Figure 12, 13, and 14 provide simulation results. From the simulation results shown below, it is obvious that the adaptive controller cannot achieve perfect tracking performance of the system output and reference trajectory.
System output y on yd in Case 3.
Control input uk and vk in Case 3.
Tracking error s(t) versus time.
As observed in simulation results above, the proposed AILC can achieve a good tracking performance and tracking errors decrease along the iteration axis, which demonstrates the validity of the proposed AILC approach in this paper.
6. Conclusions
In this paper, a new AILC scheme is proposed for a class of nonlinear time-varying systems with both unknown time-varying time-delay and unknown input dead-zone nonlinearity in the presence of disturbance running on a finite time interval repetitively. A novel representation of the dead-zone output is given. Using appropriate Lyapunov-Krasovskii functional in the Lyapunov function candidate, the uncertainties from unknown time-varying delays are removed such that control law is delay-independent. The identical initial condition for ILC is relaxed by introducing the boundary layer. The hyperbolic tangent function is employed to avoid the possible singularity problem. Theoretical analysis by constructing Lyapunov-like CEF has shown that the tracking errors converge to a small residual domain around the origin as iteration goes to infinity. At the same time, all the closed-loop signals remain bounded. Simulation results have been provided to demonstrate the effectiveness the proposed control scheme.
AppendixProof of Lemma 13.
For convenience in expression, denote x=sk(t)/η(t). We rewrite inequality (32) as
(A.1)1b<tanh2(x)=(ex-e-xex+e-x)2=1-(2ex+e-x)2.
Noting the fact that ex and e-x are positive, it follows from (A.1) that
(A.2)ex+e-x>2bb-1.
Then, we can obtain
(A.3)e2x-2bb-1ex+1>0.
Solving the quadratic inequality (A.3), we can have
(A.4)0<ex<b(b-1)-1(b-1)orex>b(b-1)+1(b-1).
On the other hand, from |sk(t)|>mη(t), we know that
(A.5)x<-morx>m
which implies
(A.6)0<ex<1b/(b-1)+1/(b-1)=b(b-1)-1(b-1)orex>b(b-1)+1(b-1).
Obviously, from the homology of (A.4) and (A.6), we know that Lemma 13 holds.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
ArimotoS.KawamruaS.MiyazakiF.Bettering operation of robots by learning1984112123140OhS.BienZ.SuhI. H.An iterative learning control method with application for the robot manipulator198845508514KucT. Y.NamK.LeeJ. S.An iterative learning control of robot manipulators199176835841OwensD. H.Universal iterative learning control using adaptive high-gain feedback19937538338810.1002/acs.4480070507MR1256810ZBL0796.93069JangT.-J.ChoiC.-H.AhnH.-S.Iterative learning control in feedback systems199531224324810.1016/0005-1098(94)00064-PMR1315131ZBL0830.93034ChienC.-J.LiuJ.-S.A P-type iterative learning controller for robust output tracking of nonlinear time-varying systems199664231933410.1080/00207179608921630MR1653756ZBL0863.93040MoonJ. H.DohT. Y.ChungM. J.A robust approach to iterative learning control design for uncertain systems199834810011004WangD.Convergence and robustness of discrete time nonlinear systems with iterative learning control1998341114451448SaabS. S.On a discrete-time stochastic learning control algorithm20014681333133610.1109/9.940946MR1847346SunM.WangD.Iterative learning control with initial rectifying action20023871177118210.1016/S0005-1098(02)00003-1MR2099405ChenH.-F.FangH.-T.Output tracking for nonlinear stochastic systems by iterative learning control200449458358810.1109/TAC.2004.825613MR2049817MadadyA.PID type iterative learning control with optimal gains200862194203FrenchM.RogersE.Non-linear iterative learning by an adaptive Lyapunov technique2000731084085010.1080/002071700405824MR1781971ChoiJ. Y.LeeJ. S.Adaptive iterative learning control of uncertain robotic systems20001472217223XuJ.-X.ViswanathanB.Adaptive robust iterative learning control with dead zone scheme2000361919910.1016/S0005-1098(99)00100-4MR1833676ZBL0939.93018XuJ.-X.TanY.A composite energy function-based learning control approach for nonlinear systems with time-varying parametric uncertainties200247111940194510.1109/TAC.2002.804460MR1937714TianY.-P.YuX.Robust learning control for a class of nonlinear systems with periodic and aperiodic uncertainties200339111957196610.1016/S0005-1098(03)00205-XMR2142831ZBL1051.93039TayebiA.Adaptive iterative learning control for robot manipulators20044071195120310.1016/j.automatica.2004.01.026MR2148314ZBL1051.93038XuJ.-X.TanY.LeeT.-H.Iterative learning control design based on composite energy function with input saturation20044081371137710.1016/j.automatica.2004.01.029MR2153050ZBL1077.93057XuJ. X.XuJ.On iterative learning from different tracking tasks in the presence of time-varying uncertainties2004341589597SunM. X.GeS. S.Adaptive repetitive control for a class of nonlinearly parametrized systems200651101684168810.1109/TAC.2006.883028MR2262072ChiR. H.HouZ. S.XuJ. X.Adaptive ILC for a class of discrete-time systems with iteration-varying trajectory and random initial condition20084482207221310.1016/j.automatica.2007.12.004MR2531355ChienC. J.HsuC. T.YaoC. Y.Fuzzy system-based adaptive iterative learning control for nonlinear plants with initial state errors2004125724732TayebiA.ChienC.-J.A unified adaptive iterative learning control framework for uncertain nonlinear systems200752101907191310.1109/TAC.2007.906215MR2355799ChienC. J.A combined adaptive law for fuzzy iterative learning control of nonlinear systems with varying control tasks20081614051WangY.-C.ChienC.-J.Decentralized adaptive fuzzy neural iterative learning control for nonaffine nonlinear interconnected systems20111319410610.1002/asjc.299MR2809540ZBL1248.93104ChenW.LiJ. M.LiJ.Practical adaptive iterative learning control framework based on robust adaptive approach2011131859310.1002/asjc.275MR2809539ZBL1248.93104GeS. S.HongF.TeeT. H.Robust adaptive control of nonlinear systems with unknown time delays200541121181119010.1016/j.automatica.2005.01.011MR2160117ZBL1078.93046HuaC. C.FengG.GuanX. P.Robust controller design of a class of nonlinear time delay systems via backstepping method200844256757310.1016/j.automatica.2007.06.008MR2530807ZBL1283.93099ChenW. S.JiaoL. C.LiJ.LiR. H.Adaptive NN backstepping output-feedback control for stochastic nonlinear strict-feedback systems with time-varying delays2009403112LiuZ.-F.CaiC.-X.ZouY.Switching signal design for exponential stability of uncertain discrete-time switched time-delay systems2013201311416292MR310892110.1155/2013/416292ZhouB.GaoH. J.LinZ. L.DuanG.-R.Stabilization of linear systems with distributed input delay and input saturation201248571272410.1016/j.automatica.2012.02.007MR2912794ZBL06072646ZhouB.LinZ. L.DuanG.-R.Truncated predictor feedback for linear systems with long time-varying input delays201248102387239910.1016/j.automatica.2012.06.032MR2961137ZBL1271.93123ZhouB.LiZ.-Y.LinZ. L.Observer based output feedback control of linear systems with input and output delays20134972039205210.1016/j.automatica.2013.03.031MR3063059ZhouB.2014Berlin, GermanySpringerChenY.GongZ.WenC.Analysis of a high-order iterative learning control algorithm for uncertain nonlinear systems with state delays199834334535310.1016/S0005-1098(97)00196-9MR1623057PrakK. H.BienZ.HwangD. H.Design of an iterative learning controller for a class of linear dynamic systems with time delay19981456507512SunM. X.WangD. W.Initial condition issues on iterative learning control for non-linear systems with time delay200132111365137510.1080/00207720110052021MR1958598ChenW. S.Adaptive learning control of linear systems with completely unknown time delays200962177185ChenW. S.ZhangL.Adaptive iterative learning control for nonlinearly parameterized systems with unknown time-varying delays201082177186ChenW. S.WangY. L.LiJ. M.Adaptive learning control for nonlinearly parameterized systems with periodically time-varying delays200834121556156010.3724/SP.J.1004.2008.01556MR2502867TaoG.KokotovićP. V.Adaptive control of plants with unknown dead-zones1994391596810.1109/9.273339MR1258675ZBL0796.93070WangX. S.HongH.SuC. Y.Model reference adaptive control of continuous-time systems with an unknown input dead-zone20031503261266TangX. D.TaoG.SureshM. J.Adaptive actuator failure compensation for parametric strict feedback systems and an aircraft application200339111975198210.1016/S0005-1098(03)00219-XMR2142833ZBL1048.93043WangX.-S.SuC.-Y.HongH.Robust adaptive control of a class of nonlinear systems with unknown dead-zone200440340741310.1016/j.automatica.2003.10.021MR2145268ZBL1036.93036ZhangT. P.GeS. S.Adaptive neural control of MIMO nonlinear state time-varying delay systems with unknown dead-zones and gain signs20074361021103310.1016/j.automatica.2006.12.014MR2306749ZBL1282.93152ZhangT. P.GeS. S.Adaptive dynamic surface control of nonlinear systems with unknown dead zone in pure feedback form20084471895190310.1016/j.automatica.2007.11.025MR2528143ZBL1149.93322XuJ.-X.XuJ.LeeT. H.Iterative learning control for systems with input deadzone20055091455145910.1109/TAC.2005.854658MR2164454GeS. S.TeeK. P.Approximation-based control of nonlinear MIMO time-delay systems2007431314310.1016/j.automatica.2006.08.003MR2266767ZBL1137.93042IoannouP. A.SunJ.1995Englewood Cliffs, NJ, USAPrentice-Hall