Robust controller design by linear programming with application to a double-axis positioning system
Introduction
Many controller design methods are based on optimization techniques. In early works, the main interest was to find an analytical solution to the optimization problem. Recently, with new progress in numerical methods to solve convex optimization problems, new approaches for controller design with convex objectives and constraints have been developed. These methods usually lead to controllers of at least the same order as the plant model. However, design of restricted-order controllers (like PID controller) leads to non-convex optimization problems in the controller parameter space.
Nowadays, PID controllers are still extensively used in industrial applications and a lot of methods have been proposed in the literature to simply tune the controller parameters. optimization of fixed-order controllers have been the subject of many research works (Malan et al., 1994; Grigoriadis & Skelton, 1996) which adopt nonlinear and non-convex algorithms. These methods require generally heavy computations and do not guarantee the global optimum to be achieved. Moreover, it is generally difficult to tune the weighting filters automatically. Industrial users prefer classical specifications like gain and phase margins, crossover frequency, maximum of the sensitivity and complementary sensitivity functions and good set point and disturbance rejection responses. A combination of time and frequency-domain specifications makes the optimization problem more complicated. A model-based method for optimizing the parameters of a PID controller with a frequency-domain criterion and closed-loop specifications is proposed in Harris and Mellichamp (1985).
The objective function is minimized with a version of the simplex method. Schei (1994) proposes a non-convex constrained optimization method to maximize the controller gain in low frequencies with constraints on the maximum of the sensitivity and complementary sensitivity functions. Åström et al. (1998) show that the maximum of the sensitivity function is an appropriate design variable and together with optimization of load disturbance rejection and good choice of set point weight can give generally very high performances for PI controllers. An algorithm to find a local minimum of the criterion is proposed. Panagopoulos et al. (2002) extend this method to design PID controllers. A numerical solution to this non-convex problem is also developed in Hwang and Hsiao (2002).
Convex optimization approaches to fixed-order controller design are rather limited. In Grassi and Tsakalis (1996) a convex optimization method for PID controller tuning by open-loop shaping in frequency domain is proposed. The infinity-norm of the difference between the desired open-loop transfer function and the achieved one is minimized. This method, however, needs the desired open-loop transfer function to be defined and it cannot be applied to the case of multi-model uncertainty. Ho et al. (1997) show that all stabilizing PID controllers with fixed proportional gain can be found by resolving a linear programming problem for the derivative and integral gains. Blanchini et al. (2004) go further by showing that, given the value of the proportional gain, the region of the plane defined by the derivative and integral gains, where a considered constraint is satisfied, consists of the union of disjoint convex sets. However, both methods have the drawback of fixing the proportional gain a priori. In Keel and Bhattacharyya (1997) a PID controller design based on linear programming for pole placement and model matching problem is proposed. A serious difficulty in this approach is to specify the desired closed-loop poles and desired closed-loop transfer function. Recently, a new approach based on the generalized Kalman–Yakoubovic–Popov lemma has been proposed to tune the linearly parameterized controllers in the Nyquist diagram (Hara et al., 2006). The idea is to define several convex regions in the complex plane and design the controller such that in each frequency interval the open-loop transfer function lies in one of the regions. However, this method seems to be too complex for industrial applications.
In this paper, a loop shaping method in the Nyquist diagram is proposed. A new stability margin is defined which guarantees a lower bound for the gain, phase and modulus margins (the inverse of the maximum of the sensitivity function). The main property of the new margin is that a constraint on this margin is linear with respect to the parameters of linearly parameterized controllers. Therefore, optimizing load disturbance rejection with constraint on this margin leads to a linear constrained optimization problem which can be solved by linear programming. A lower bound for the crossover frequency is also defined which also leads to a linear constraint for the optimization problem. With this constraint, an optimization problem to maximize the new robustness margin can be solved by linear programming. The method can be applied to stable plants represented by transfer functions with pure time delay or simply by non-parametric models in the frequency domain. The robustness of the closed-loop system with respect to unmodeled dynamics is ensured with the constraint on the new linear margin, while the multi-model uncertainty can be considered easily by increasing the number of constraints. The proposed method can be used for PID controllers as well as higher-order linearly parameterized controllers in discrete or continuous time. Standard linear optimization solvers can be used to find the global optimal controller. The proposed method is applied to a double-axis linear permanent magnet synchronous motor (LPMSM). This type of motors can be used for wafer fabrication and inspection and other high precision positioning systems. A generalized synchronization control for such systems has been proposed in Xiao et al. (2005).
This paper is organized as follows: in Section 2 the class of models, controllers and the control objectives are defined. Section 3 introduces a new linear stability margin, a lower bound on the crossover frequency and presents the linear optimization problem. Simulation results are given in Section 4 and experimental results in Section 5. Finally, Section 6 gives some concluding remarks.
Section snippets
Plant model
The class of linear time-invariant SISO systems with no pole in the right half plane is considered. It is assumed that a set of non-parametric models in the frequency domain is available. This set can be obtained either by spectral analysis from several identification experiments at different operating points or from parametric models in the form of rational transfer functions with pure time delay. Suppose that the dynamics of the system can be captured by a sufficiently large finite number of
Controller design by linear programming
The classical robustness indicators like gain, phase and modulus margins as well as the performance indicator, the crossover frequency , are nonlinear functions of the controller parameters. The optimization methods with constraints on these values lead to non-convex optimization problems and cannot be solved efficiently. In this section a new stability margin together with a lower bound for the crossover frequency will be defined which leads to linear constraints for the optimization
Simulation results
The design method is tested on two different examples taken from Panagopoulos et al. (2002) to illustrate its properties.
Experimental results
The proposed method is applied to a LPMSM. The aim is to control the position of such a system, which is shown in Fig. 9. LPMSMs have high acceleration and deceleration capabilities, high mechanical stiffness, reduced friction and no mechanical transmission. They, therefore, do not suffer from backlash and thus allow very high positioning accuracy to be achieved. In this application, the dynamics of each axis depend on the position of the two axes. For example, it is clear that the dynamics of
Conclusions
Robust fixed-order controller design is formulated as a linear optimization problem. The proposed method is based on frequency loop shaping in the Nyquist diagram. The classical robustness and performance specifications are represented as linear constraints in the Nyquist diagram. There are only a few design variables which are directly related to the robustness (linear margin and ) and performance (lower approximation of the crossover frequency ) of the closed-loop system. The control
Acknowledgments
The authors acknowledge the collaborations of Ralph Coleman, Michel Mathia and Vincent Very from the company ETEL for real-time experiments.
References (15)
- et al.
Design of PI controllers based on non-convex optimization
Automatica
(1998) - et al.
Low-order control design for LMI problems using alternating projection methods
Automatica
(1996) - et al.
Solution of a non-convex optimization arising in PI/PID control design
Automatica
(2002) Automatic tuning of PID controllers based on transfer function estimation
Automatica
(1994)- et al.
Generalized synchronization control of multi-axis motion systems
Control Engineering Practice
(2005) - et al.
PID controllers: Theory, design, and tuning
(1995) - et al.
Characterization of PID and lead/lag compensators satisfying given specifications
IEEE Transactions on Automatic Control
(2004)
Cited by (67)
Optimized four-parameter PID controller for AVR systems with respect to robustness
2022, International Journal of Electrical Power and Energy SystemsFrequency-Domain Data-Driven Controller Synthesis for Unstable LPV Systems
2021, IFAC-PapersOnLineSISO approaches for linear programming based methods for tuning decentralized PID controllers
2020, Journal of Process ControlCitation Excerpt :The proposedmethods for PID tuning of MIMO systems developed in this work heavily rely on such results, and therefore the authors believe that their adequate understanding is imperative to interpreting the results presented in this work correctly, as well as for successfully applying the proposed methods. The PID controller design method for SISO systems of Karimi et al. [5] computes the controller parameters via a convex optimization problem that minimizes the disturbance effects subject to stability margin constraints. In this section, the details of this technique are presented as they are important to understand the PID controller design for MIMO systems proposed in this work.
Accounting for dynamics in self-optimizing control
2019, Journal of Process ControlPerformance enhancement of hard disk drives through data-driven control design and population clustering
2019, Precision EngineeringCitation Excerpt :To design controllers for non-parametric (data-driven) models, two design frameworks have been pursued. The most common approach is to formulate a convex optimization problem to minimize a certain cost function such as the L2/L∞ norm of the sensitivity transfer function error from a desired one under a PID control law [13], or robust controller design in the Nyquist diagram using linear/quadratic programming and H∞ controller design [14–16]. The second approach is to formulate a non-convex multi-objective optimization problem, including all the required robustness and performance objectives.