Professional Documents
Culture Documents
Introduction
Introduction
INTRODUCTION
Many practical systems call for an improvement in transient performance along with the steadystate accuracy. For example, many electro-mechanical, robotics and power converter systems require a quick response without any overshoot. It is a well-understood fact that a low overshoot can be achieved at the cost of a high settling time. However, a short settling time is also necessary for a quick response. Thus, most of the design schemes make a tradeoff between these two transient performance indices and the damping ratio is chosen as a fixed number. Notable exceptions exist, of course, in the proposed a seminal idea of composite non-linear feedback (CNF) for a class of second-order systems subject to actuator saturation. Composite non-linear control uses a variable-damping ratio to achieve high performance. Initially, it keeps damping ratio to a low value to ensure quick response and as the output approaches the set point the damping ratio is increased and thus overshoot is avoided. Subsequently, CNF controller was extended for general higher-order single-input and single-output (SISO) and multiple-input and multiple-output (MIMO) systems in for the state feedback and output-feedback cases. The second important requirement is robustness towards uncertainty, because most of the physical systems are represented by approximate mathematical models to ease analysis. However, model mismatches and discrepancies are inevitable between the actual plant and the mathematical model used for the controller design. These mismatches along with external unknown disturbances affect the performance of the system. To tackle the effect of parameter variations and external disturbances, many robust control techniques have been developed. One of the robust control techniques is variable structure control (VSC); in VSC, the structure of closed-loop system is changed as per some decision rule; this rule is called the switching function. The robustness property of VSC is because of forcing of sliding mode and eventually sliding mode became a principal operating mode in VSC. The resulting control system design is termed as the sliding mode control (SMC); recently many successful practical applications of SMC have established the importance of sliding mode theory which has mainly been developed in the last three decades. This fact is also witnessed by many special issues of learned journals focusing on SMC. To relax the need for measuring the entire state vector, an output feedbackbased sliding mode concept is also proposed in which widens the scope of SMC. In SMC, the sliding surface decides closed-loop dynamics; therefore it should be designed such that it addresses all requirements. As mentioned earlier, high performance in an uncertain environment is one of the key requirements. To enhance the performance of the system with SMC algorithms, a time-varying switching surface is proposed by many researchers in .In a strategy based on fuzzy logic is devised to change the parameters of switching surfaces of higher-order systems. Some researchers proposed non-linear surfaces to improve the performance. The variable-damping ratio improves the system performance significantly. This paper proposes a method to design a non-linear sliding surface for a class of non-linear uncertain system. A non linear sliding surface is designed by using the principle of a variable-damping concept. Using a non-linear sliding surface, the damping ratio of a system can be changed from
1|P a ge
its initial low value to a final high value. The initial low value of damping ratio results in a quick response and the later high damping avoids overshoot. Thus, the proposed surface ascertains the reduction in settling time without any overshoot. Furthermore, the systems damping ratio changes continuously as per the chosen function. The proposed approach inherits the robustness of SMC and delivers high performance owing to change of damping ratio by means of the non linear sliding surface. During sliding mode, because of the order reduction, the system response is unaffected by m poles. For the systems of order higher than two, the damping ratio is specified by considering the contribution of dominant poles. However, non-dominant poles always affect the systems response to some extent depending on their relative locations with respect to the dominant poles. Because of the order reduction property of SMC, m non-dominant poles will not contribute to the system response and thus, the performance specifications can be met more closely. The proposed nonlinear sliding surface achieves high performance and robustness unlike a sliding surface designed by assigning eigen values or by minimizing a quadratic index that normally lead to a linear-sliding surface. To ease the synthesis of the proposed non-linear surface a linear matrix inequality (LMI)-based algorithm is also proposed.
1.1. Conclusion
y y y In this work, we propose the following. A non-linear sliding surface for a class of non-linear uncertain SISO system to achieve low overshoot and short settling time simultaneously. Existence of sliding mode around the proposed non-linear sliding surface and a new control law. An LMI-based technique to ease the synthesis of the proposed non-linear surface. The next section discusses the sliding mode control technique.
2|P a ge
2.2.
Background
3|P a ge
4|P a ge
Here, His a strict positive constant, taken to be the bandwidth of the system, and xd (t) is the desired state. The problem of tracking the n-dimensional vector xd (t ) can be replaced by a first-order stabilization problem in s. s(x; t ) verifying (2.2) is referred to as a sliding surface, and the systems behavior once on the surface is called sliding mode or sliding regime. From (2.2) the expression of s contains , we only need to differentiate s once for the input u to appear. Furthermore, bounds on s can be directly translated into bounds on the tracking error vector , and therefore the scalar s represents a true measure of tracking performance. The corresponding transformations of performance measures assuming is: (2.3) i=0, 1,2n-1. The simplified, 1st-order problem of keeping the scalar s at zero can now be achieved by choosing the control law u of (1) such that outside of S(t) where Lis a strictly positive constant. Condition (2.4) states that the squared distance to the surface, as measured by s2 decreases along all system trajectories. Thus, it constrains trajectories to point towards the surface s(t). In particular, once on the surface, the system trajectories remain on the surface. In other words, satisfying the sliding condition makes the surface an invariant set (a set for which any trajectory starting from an initial condition within the set remains in the set for all future and past times). Furthermore (2.4) also implies that some disturbances or dynamic uncertainties can be tolerated while still keeping the surface an invariant set.
5|P a ge
Finally, satisfying (5.2) guarantees that if x(t=0) is actually off xd(t=0), the surface S(t) will be reached in a finite time smaller than . Assume for instance that s(t=0)>0, and let treach be the time required to hit the surface s=0. Integrating (4) between t=0 and treach leads to 0-s(t=0)=s(t= treach)-s(t=0) e-L( treach -0), which implies that treach es(t !0) / L The similar result starting with s(t=0)>0 can be obtained as treach es(t = 0) /L Starting from any initial condition, the state trajectory reaches the time-varying surface in a finite time smaller than s(t = 0) /L, and then slides along the surface towards x d ( t ) exponentially, with a time-constant equal to 1 / P. In summary, the idea is to use a well-behaved function of the tracking error, s, according to (2.2), and then select the feedback control law u in (1) such that s2 remains characteristic of a closedloop system, despite the presence of model imprecision and of disturbances.
6|P a ge
Differentiating the above expression and substituting (2.5) we get The approximation of control law to achieve is To account for the uncertainty in f while satisfying the sliding condition
take the control law as: By choosing k(x,t) large enough, such as
Hence, by using (2.11), we ensure the system trajectory will take finite time to reach the surface S(t), after which the errors will exponentially go to zero. From this basic example, we can see the main advantages of transforming the original tracking problem into a simple 1st-order stabilization problem in s. In first-order systems, the intuitive feedback control strategy if the error is negative, push on the positive direction; if the error is positive, push on the negative direction works. The same statement is not true in higher-order system. Now consider the second order system in the form of Where b(x,t) is bounded. If is the gain margin of the design, it can be proved that the control law with satisfies the sliding condition. The control law for higher order system can be deducted based on similar approach.
7|P a ge
frequency. The representative point then oscillates within a neighborhood of the switching surface. This oscillation, called chattering, is illustrated on figure 2.2.
Figure 2.2: Chattering as a result of imperfect control switching. Control laws which satisfying sliding condition (2.4) and lead to perfect tracking in the face of model uncertainty, are discontinuous across the surface S(t), thus causing control chattering. Chattering is undesirable, since it involves extremely high control activity, and furthermore may excite high-frequency dynamics neglected in the course of modeling. Chattering must be reduced (eliminated) for the controller to perform properly. This can be achieved by smoothing out the control discontinuity in a thin boundary layer neighboring the switching surface.
2.3 Conclusion
Due to its robustness properties, sliding mode controller can solve two major design difficulties involved in the design of a braking control algorithm: 1. The vehicle system is highly nonlinear with time-varying parameters and uncertainties; 2. The performance of the system depends strongly on the knowledge of the tire/road surface condition; For the class of systems to which it applies, sliding controller design provides a systematic approach to the problem of maintaining stability and consistent performance in the face of modeling imprecision. For wheel slip control system, the vehicle and brake system are highly nonlinear and timevarying systems. That makes sliding mode controller ideal candidate for the application.
8|P a ge
Here , and are uncertain smooth functions satisfying the classical condition for the existence and uniqueness. With the relative degree n, the above non-linear system, using appropriate transformation, can be transformed into the following form T where x(t) = [x1 x2 xn] represents the state vector, and d2(x, t) is an uncertain smooth function that satisfies the classical condition for existence and uniqueness. Furthermore, the function satisfies Where Q > 0 and Q1 and Q2 are positive constants. The above definition relaxes the need of precise modeling of the plant. The above non-linear system is in a Brunowsky canonical form. B2 is a non-zero scalar. The aforementioned system can be represented in the so-called regular as Where Where z1(t)
n-1
Let the sliding surface be defined as follows
9|P a ge
F is chosen such that (A11 A12F) has stable eigenvalues and the dominant poles have a low damping ratio. ( y) is a non positive differentiable function in y, used to change the damping ratio. P is a positive definite matrix that is chosen based on the desired final damping ratio 2, and also
for some positive definite matrix W. Using this non-linear sliding surface, the damping ratio of the system is increased from its initial value 1 to the final value 2. The formulation to calculate the matrix P, based on the required 2, is discussed later in this section.
where y0= y(0) and is used as a tuning parameter. This parameter contributes to decide the final damping ratio along with the matrix P. It should be noted that the choice of ( y) is not unique and any function with the abovementioned properties can be used. Another possible choice is as follows Where is a positive constant that should have a large value to ensure a small initial value of .
From (3.4) and (3.12), the system equation during sliding mode becomes
The stability of sliding mode can be proved by showing that the subsystem (3.13) is stable. The stability of the (3.13) can be proved using the following theorem. Theorem 3.1: If is stable and is defined by (3.10) or (3.11) then the subsystem (3.13) is stable. Proof: Let a Lyapunov function for the system in (3.13) be defined as Then it follows that
Therefore we have where M := PA12 R(n1). As PA12AT12P = MMT 0 and the function ( y) is negative by definition, therefore matrix 2 ( y)MMT is negative semi-definite. The matrix W is negative definite and the addition of a negative definite and a negative semi-definite matrix always results in a negative definite matrix. Therefore we can write (z) < 0, which completes the proof.
Let us assume, without loss of generality, a third-order uncertain system described by (3.5). A suitable F can be chosen so that the matrix A11- A12F becomes Hurwitz and has a small damping ratio. Consider this system to be in the canonical form as described by (3.5) and thus
Also define
where a1 and a2 are positive constants. Let us define the positive-definite matrices P and W which satisfy (3.9) as
and
By using the above values of matrices, (3.13) can be rewritten and simplified as
The eigen values of the above closed-loop system can be computed as
To obtain the above equations in terms of elements of matrix W, the following formulae can be derived easily by using (3.9) as
As discussed earlier, the function changes from its initial value zero to some negative value. As the function changes, this results in the change in location of closed loop poles. Now, it can be easily seen that, as and , the closed-loop poles in (3.19) can be written as
It can be concluded, for this general third-order system, that during sliding mode, as the function changes from its initial value to , the closed-loop systems damping ratio increases to . To analyze a higher-order system a reduced order model, which is formed by the use of only
12 | P a g e
dominant poles, can be used. Similarly, when the function changes from 0 to , where some finite value, the corresponding increase in the damping ratio can also be proved.
is
To achieve the desired damping ratio, a matrix P is needed which satisfies (3.24). To ensure stability of the closed-loop system, the matrix P also needs to satisfy (3.9). However, it may not always be possible for the matrix P to satisfy constraints (3.9) and (3.24) simultaneously. Therefore one may relax constraint (3.24) as follows for a sufficiently small , where
It may be noted that the above non-linear inequality can be converted to a linear inequality by using Schur complement as follows
Therefore we can cast the optimization problem as follows Choose P to minimize such that
13 | P a g e
The above LMIs can be easily solved to minimize by using many commercial computational tools. It should be noted that the freedom to choose the parameter can also be used to minimize the objective function. This approach automates the tuning procedure and reduces the number of manual iterations required for tuning.
3.4 Conclusion
This section discussed the design of a non-linear sliding surface for a class of non-linear uncertain SISO system. The non-linear function (y) is introduced to change the systems closed-loop damping ratio from its initial low value to a final high value as the output varies from its initial value and approaches the set point. The stability of the sliding surface is discussed. LMI based synthesis of the non-linear surface is also explained. The section deals with the existence of sliding surface.
14 | P a g e
enforces the trajectory of (3.4) to move from any initial condition to the sliding surface in finite time and remain on it thereafter. In the above control law, the scalar Q is chosen from the maximum bound of the uncertainty that satisfies (3.3) and this implies K is a positive constant. Proof: For a quadratic function
15 | P a g e
Considering (4.2), it is straightforward to verify, for some . This ensures finite time convergence of s and thus it completes the proof. It should be noted that using the control law (4.1) the state trajectory will hit the surface in finite time. After hitting the sliding surface, the trajectory will slide along the surface and the system is invariant to parameter variations and external disturbances satisfying the matching condition.
4.1 Conclusion
The control law is chosen in such a way that from any initial condition, the way that from any initial condition, the system trajectory is attracted towards the sliding surface and then slides along the surface. The existence of sliding mode on the surface is proved.
16 | P a g e
Output equation is Step 1: Design the non-linear sliding surface. Non-linear sliding surface is composed of a constant and a non-linear term. Initially the non-linear term is zero, therefore the constant term decides the initial damping ratio (z1) and the settling time. Sliding surface can be given by the following equation. As per the discussion in Section 3, let c2 = 1, and c1 is made of two components linear and nonlinear. F is designed for the low damping ratio. For initial settling time ts1=2s and initial damping ratio . Let =50 and 1=0.4, matrix F can be found from pole placement approach as F= solving Lyapunov equation with
For P gives
Function
17 | P a g e
Step 2: Design of control law. Control law can be given as Following values are chosen k=20, Q=1.01, x2=[0.1 0 0]T. The value of Q is chosen by considering maximum bound of uncertainty which is ( d2)max =1. 5.1. Design of Sliding Surface Performance of the proposed non-linear sliding surface is compared with the controller designed with different linear-sliding surfaces. System output response is plotted for different linear surfaces and the proposed non-linear surface. When non-linear sliding surface is used, poles of the closed-loop system changes as output approaches to the origin. Following three different linear-sliding surfaces are designed 1. Case 1: Linear surface 1 with =0.7, ts=2s. 2. Case 2: Linear surface 2 with =0.6, ts=2s. 3. Case 3: Linear surface 3 with =0.8, ts=2s. Fixed sliding surface with given settling time and damping ratio is designed by pole placement approach. In the above equation Fi is computed by pole-placement approach for each of the three cases mentioned above with corresponding damping ratio and settling time. Control law for each of the three cases mentioned above is given as follows In the above control law, parameters Q and k are the same as (5.4). The system is simulated with the same initial condition and disturbance for different linear-sliding surfaces.
18 | P a g e
because of non-linear sliding surface gain increases as state reaches nearer to the origin and thus the settling time improves.
19 | P a g e
Fig.5. 3 Plot of input with different linear surfaces and non-linear surface.
5.3 Conclusion
Ship roll stabilization was presented to illustrate the proposed method. Simulation results were presented for different damping ratio sliding surfaces. These results were compared with the proposed non-linear sliding surfaces.
20 | P a g e
6. CONCLUSION
The design of non-linear sliding surface is proposed to improve the performance. The existence of sliding mode on such a surface is proved. Ship roll stabilization example clearly shows that the performance significantly improves. The proposed algorithm is able to achieve low overshoot and short settling time simultaneously for a class of uncertain non-linear system. Robustness and high performance both can be achieved with the proposed scheme. Disturbance rejection property of the proposed method is also shown. Also, we note that it is straightforward to extend our results for MIMO case.
21 | P a g e
REFERENCES
1. D.Fulwani, B.Bandyopadhyay, L.Fridman Non-linear sliding surface: towards high performance robust control IET Control Theory Application 2012 , Vol.6 Iss.2, pp.235-242. 2. Weiyao Lan, Ben M. Chen,Yingjie He On improvement of transient performance in tracking control for a class of non-linear system with input saturation W. Lan et al. / Systems & Control Letters 55 (2006) 132 138. 3. S. Tokat, I. Eksin, M. Guzelkaya Sliding mode control using a non-linear time varying sliding surface Proceedings of the 10th Mediterranean Conference on Control and Automation - MED2002 Lisbon, Portugal, July 9-12, 2002. 4. K. Peng, G. Cheng, B.M. Chen and T.H. Lee Improvement of transient performance in tracking control for discrete-time systems with input saturation and disturbances IET Control Theory Appl., Vol. 1, No. 1, January 2007.
22 | P a g e