Frequency domain precision analysis and design of sliding mode observers

https://doi.org/10.1016/j.jfranklin.2009.12.003Get rights and content

Abstract

Estimation precision and bandwidth of sliding mode (SM) observers are analyzed in the frequency domain for different settings of the observer design parameters. It was shown previously that the SM observer could be analyzed as a relay feedback–feedforward system. It is feedback with respect to the measured variable of the system being observed, and feedforward with respect to the control applied to the system being observed. This approach is now further extended to analysis of effects of design parameter change on observer performance. An example of SM observer design for estimation of DC motor speed from the measurements of armature current is considered in the paper. The input–output properties of observer dynamics are analyzed with the use of the locus of a perturbed relay system (LPRS) method.

Introduction

The idea of using a dynamical system, which is called observer, to obtain estimates of the system states from measurable system variables was proposed by Luenberger [1]. The observer dynamics are driven by the control and by the difference between the output of the observer and the output of the plant. In SM observers, this difference is maintained equal to zero by means of SM organized in the observer loop. The control should be designed to provide the existence of SM in the observer dynamical system. SM observers were analyzed in a number of publications (see, for example, respective chapters of textbooks [2], [3] and recent tutorials [4], [5]).

However, only ideal SM in the observer dynamical system was analyzed in [2], [3], [4], [5]. To the best of our knowledge, only in [6] for the first time the mechanism of generation of the observation error was found and analyzed through a frequency-domain approach. The proposed approach was based on the locus of a perturbed relay system (LPRS) method [7] and the method of analysis of SM systems with parasitic dynamics [8]. Further development of the approach of [6] was presented in [9] for second-order SM observers. Publication [10] provides some results that can be considered as experimental verifications of the approach of [6], [9]. However, in publication [6], the proposed frequency-domain approach was just outlined. The present paper further develops this approach extending it to the observer design problem and providing an example of its application to DC motor speed observation. This new development is aimed at reaching out to the area of engineering design of SM observers.

The objective of the present paper is, therefore, to extend the approach of [6] to design-related problems, in particular, to investigation of precision dependence on observer design parameters (execution period of the algorithm and values of the observer gain matrix). We aim to show that the observation precision has complex dependence on the SM algorithm parameters. We investigate that dependence and provide a model that allows for computing the observation error. The paper is organized as follows. At first, the problem formulation is considered. Then a frequency-domain approach to SM observer analysis is presented, and a model that provides precision of observation is given. After that, considering the example of DC motor speed observation, an investigation of the dependence of the observation precision on the values of the algorithm design parameters is done, and the design is outlined.

Section snippets

Problem formulation

Consider an n-dimensional version of the observer proposed in [2]. Let the linear plant, the states of which are supposed to be observed, be the nth order dynamical system:x˙=Ax+Buy=Cxwhere xRn is the state vector, yR1 is the measurable system output, ARn×n, BRn×1, and CRn are the state matrix, the input matrix, and the output matrix of the plant, respectively. The pair (C,A) is assumed to be observable.

The SM observer can be designed in the same form as the original system (1) and (2)

The concepts of the locus of a perturbed relay system (LPRS) approach

In [7], [8], the LPRS was introduced as a method of analysis and design of relay servo systems having a linear plant (Fig. 2). Let us call the part of the relay servo system that is given by the linear differential equations the linear part. With respect to the SM observer, the linear part will be the one given by Eqs. (3), (4).

The LPRS was defined as a complex function J(ω) of the frequency ω as follows:J(ω)=12limf00σ0u0+jπ4climf00y(t)|t=0where t=0 is the time of the switch of the relay

Analysis and characteristics of SM observer performance

With the representation of the SM observer as a relay servo system, we formulate performance measures of the observer. Using the LPRS method and the concept of the equivalent gain, we obtain a linear model of the plant-observer dynamics for average (on the period of chattering) motions. We characterize the precision of observation by the output error Δy=yy=σ and by the state observation errors Δx=xx. We note that the output error and state observation errors are not equal to zero [6], [9]

Selection of matrix L and observer precision

Apparently, selection of matrix L must have an effect on observer precision, which follows from formulas (12), (15), and (16). Yet, the choice of matrix L is limited by the conditions of the existence of the sliding mode in the observer dynamics and other conditions that were discussed above. We now transform the condition of the existence of the sliding mode. Assume the absence of parasitic dynamics and the autonomous mode (u(t)≡0, which leads to y(t)≡0). We transform the condition σσ˙<0 of

Example of SM observer precision analysis

Consider an example of estimation of DC motor speed and acceleration from the measurement of armature current. The motor model is given by the diagram (Fig. 5).

In Fig. 5, Ωm is a motor speed, ia is an armature current, Ea is an armature voltage (control), L is an armature inductance, R is an armature resistance, Jt is a moment of inertia of rotor and load, Bt is a friction coefficient, Ke is a back e.m.f. coefficient, and Kt is a torque constant. We transform the original model of the motor

Conclusion

Analysis of a SM observer is done above as of a relay feedback–feedforward system. The frequency-domain model of observation precision is obtained via application of the LPRS method. It is found that the dynamical performance of the SM observer, which translates into observation precision, is not ideal. Because of the existence of parasitic dynamics in the observer loop (time delay due to discrete implementation of the algorithm) there always exists a non-zero observation error—even after the

References (13)

There are more references available in the full text version of this article.

Cited by (9)

  • A real-time estimation and control scheme for induction motors based on sliding mode theory

    2014, Journal of the Franklin Institute
    Citation Excerpt :

    On the other hand in the last decade remarkable efforts have been made to reduce the number of sensors in the control systems because the sensors increase the cost and also reduce the reliability of the control system because these elements are generally expensive, delicate and difficult to install [19–22]. Recently, sliding mode observers have been used for estimating states of systems with modeling uncertainties due to their simplicity and robust performance [23–25]. In this sense a flux estimation algorithm, based on a sliding mode observer (SMO), is proposed in this paper in order to avoid the flux sensors.

  • Sliding mode optimal control for linear systems

    2012, Journal of the Franklin Institute
    Citation Excerpt :

    Promising modifications of the original sliding mode concept, such as integral sliding mode [11] and higher order sliding modes [12,3], have been developed. The sliding mode technique is actively used as well in observation and identification methods [13–18]. Application of the sliding mode method is extended even to stochastic systems [19–21] and stochastic filtering problems [22,23].

  • Sliding mode controller design for linear systems with unmeasured states

    2012, Journal of the Franklin Institute
    Citation Excerpt :

    Since the sliding mode control was invented in the beginning of 1970s (see a historical review in [1–3]), it has been applied to solve several classes of problems. For instance, the sliding mode technique is actively used in observation and identification methods [4–9]. Application of the sliding mode technique is also extended to stochastic systems [10–12] and stochastic filtering problems [13,14].

  • Launch vehicle attitude control using sliding mode control and observation techniques

    2012, Journal of the Franklin Institute
    Citation Excerpt :

    However, the robustness issues are still not fully addressed by means of these control techniques. The Sliding Mode Control (SMC) algorithm [8,9,24–28], is a Lyapunov function based technique, providing global stability thus avoiding gain scheduling. It is an attractive design because of its inherent insensitivity and robustness to unknown plant uncertainties and external disturbances.

  • Approximate mining of global closed frequent itemsets over data streams

    2011, Journal of the Franklin Institute
    Citation Excerpt :

    Later, Li et al. [12] proposed the AFIM to approximately mine the closed frequent itemsets with a limited error tolerance. In [13], Boiko uses the sliding mode observers to obtain estimates of the dynamical system states. The damped model uses the entire stream to compute the frequency with a decay factor d, which makes the recent data more important than the previous ones.

View all citing articles on Scopus
View full text