Frequency Response Function identification for multivariable motion control: Optimal experiment design with element-wise constraints☆
Introduction
High-precision positioning systems face ever increasing performance demands in terms of accuracy and productivity ([1]). These mechatronic systems, including wafer scanners, printing systems, and medical scanners, typically exhibit complex multivariable dynamics, e.g., due to flexible mechanics. Meeting the performance demands requires the employment of advanced model-based control strategies ([2]). The majority of these control techniques rely on the availability of accurate FRF models ([3], [4], [5]).
The identification of FRF models from experimental data, as opposed to first principles modeling, is considered fast, inexpensive, and accurate ([6], [7]). The resulting FRF constitutes an intermediate step towards parametric modeling and is also used directly for control design ([8]) and system diagnostics ([9]).
The quality of the identified FRF depends on the experiment ([10]) and therefore these experiments must be designed carefully. This applies especially to next-generation mechatronic systems, equipped with a large number of actuators ([11], [5]), since the number of experiments increases with the number of inputs.
Optimal Experiment Design (OED) consists in a systematic approach for the design of input signals that maximize the FRF accuracy within limited resources. The topic of OED has received much research interest in the last decades, see the overviews [10], [12], [13].
Dealing with system constraints is a necessary aspect in experiment design to achieve practical requirements such as plant-friendly operation ([14]). For many mechatronic systems such as motion stages, the constraints are related to physical limitations of specific elements of the system, e.g., a bounded power for a specific actuator ([15]).
OED for Single Input Single Output (SISO) systems is well-developed, which has led to a large number of theoretical results and design frameworks. Design methods for parametric identification of SISO systems based on numerical optimization schemes are developed in, e.g., [7, Ch. 13], [16, Ch. 6], [10], [13]. Typically, the aim is to formulate the design problem as the minimization of a convex objective function over a convex constraint set (e.g. [17], [18], [19], [20]]). Because the optimal experiment depends on the system itself, preliminary experiments may be employed to acquire the required problem data prior to the actual optimization ([21]).
For Multiple Inputs Multiple Outputs (MIMO) systems, the design problem becomes significantly more complex. For such systems, the problem involves not only the design of excitation magnitudes, but also the excitation directions ([15]). In [6, Ch 2,5], [22], [23] the strategy for dealing with multiple inputs is to choose the excitation directions orthogonal to each other. These orthogonal excitations are (-)optimal in the case of input power constraints, but not in the general case, e.g., under bounded input and output power. Numerical optimization-based approaches that allow for multivariable design beyond orthogonal excitations are developed in [24], [25]. Herein, input design problems under total input or output power constraints are considered, which are typically convex problems and hence can be solved efficiently. For many real-life mechatronic systems, however, constraints on the total power do not adequately represent the actual, often element-wise system limitations.
In OED for MIMO systems, addressing element-wise constraints in a non-conservative manner is crucial to maximize the model quality. In prior work [15], it is shown that the design problem under element-wise constraints requires solving a non-convex optimization problem, but no general computational tools are provided. In related work [26], preliminary results to solve this problem are reported.
Although many results on OED are present, fast and accurate identification of complex multivariable mechatronic systems is hampered by a lack of design approaches that can deal with element-wise constraints. The aim of this paper is to solve the OED problem for FRF identification of complex multivariable systems under element-wise power constraints.
The main contributions in this paper are:
- 1.
A framework for optimal multivariable excitation design for FRF identification of MIMO systems under element-wise power constraints,
- 2.
a Sequential Semi-Definite Relaxation (SSDR) algorithm to approximately solve the multivariable excitation design problem, including feasibility and convergence results,
- 3.
an excitation design approach based on a Relaxation and Randomization (RR) algorithm, including theoretical and practical performance bounds,
- 4.
a comparison in terms of time-efficiency and performance of both algorithms based on experimental results from a next-generation wafer stage.
In Section 2, the mechatronic system is introduced and the experiment design framework and objective is established. In Section 3, the role of multivariable excitations in identification of MIMO systems is illustrated. Section 4 presents different algorithms to solve the design problem. The experimental results are presented in Section 5, followed by the conclusions in Section 6.
denotes the set of Hermitian matrices. Subscripts denote positive (semi)-definiteness of a matrix or (semi)-positivity of a vector. represents the complex conjugate of . Operation denotes the Frobenius inner product of equal sized matrices and . The operator denotes the Kronecker product, and results in a diagonal matrix with the entries of vector on the diagonal. indicates that a sample belongs to a normal distribution with expected value and variance . The Discrete Fourier Transform (DFT) of a discrete time sampled signal is defined as with the number of samples, the sample time, and the discrete frequency index. The normalization factor of in (1) is employed instead of the conventional factor to simplify the equations in this paper.
Section snippets
System description and OED framework
In Section 2.1, a description of the wafer stage system is provided. The methods in this paper are directly applicable to many other applications, in particular mechatronic systems with multiple actuators and sensors. In Sections 2.2 Identification framework, 2.3 Optimal experiment design objective, the identification framework and experiment design objectives are presented, respectively.
Role of multivariable excitations in OED
In this section, the role of multivariable excitation design in the identification of MIMO systems is investigated. To this end, the limitations of commonly used traditional excitation techniques are pointed out. These limitations underline the necessity of multivariable approaches.
A traditional approach to performing the identification experiments is to excite a single system input at a time. Throughout, this approach will be referred to as the Single Input Multiple Outputs (SIMO) approach.
Solutions to the multivariable OED problem
This section presents methods to compute multivariable excitations. Since the OED problem () is non-convex and generally NP-hard, the main focus is on solving the problem approximately.
In Section 4.1, () is reformulated into a rank-one constrained Semi-Definite Program (SDP), which forms an intermediate yet instrumental step in solving the problem. In Section 4.2, exact solutions to two special cases are given. Section 4.3 presents a Sequential Semi-Definite Relaxation (SSDR) algorithm,
Experimental validation
The developed experiment design algorithms are validated by means of experiments on the wafers stage system described in Section 2.1. This section presents the experimental goals, procedures, and results.
Conclusions
The identification of FRFs of mechatronic systems can be made more accurate and/or faster through the multivariable experiment design framework presented in this paper. The developed methods improve upon traditional excitation design techniques by exploiting the design freedom in the multivariable excitation signals, while efficiently dealing with the element-wise constraints of the system. This improvement is supported by experimental results from a 7 × 8 wafer stage, which show a
CRediT authorship contribution statement
Nic Dirkx: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Visualization, Writing - original draft, Writing - review & editing. Jeroen van de Wijdeven: Methodology, Software, Validation, Writing - review & editing. Tom Oomen: Supervision, Methodology, Conceptualization, Writing - review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Nic Dirkx received the M.Sc. degree from the Eindhoven University of Technology, Eindhoven, The Netherlands, in 2011. Since 2012, he is a mechatronics design engineer at ASML, Veldhoven, The Netherlands. As of 2018, he is also pursuing the Ph.D. degree in the Control Systems Technology group within the department of Mechanical Engineering at TU/e. His research interests are in the field of system identification, precision motion control, and mechatronic systems.
References (47)
- et al.
Multivariable feedback control design for high-precision wafer stage motion
Control Eng Pract
(2002) From experiment design to closed-loop control
Automatica
(2005)- et al.
Least costly identification experiment for control
Automatica
(2006) - et al.
Optimal input design using linear matrix inequalities
IFAC Proc Vol
(2000) - et al.
Accurate estimation of multivariable frequency response functions
IFAC Proc Vol
(1996) - et al.
Identification for control of multivariable systems: Controller validation and experiment design via LMIs
Automatica
(2008) Structured low-rank approximation and its applications
Automatica
(2008)- et al.
A rank-constrained optimization approach: Application to factor analysis
IFAC Proc Vol
(2014) - et al.
Rank-constrained optimization and its applications
Automatica
(2017) - et al.
The design of high performance mechatronics
(2014)
Multivariable feedback control: Analysis and design
Advanced motion control: An industrial perspective
Eur J Control
Advanced motion control for precision mechatronics: Control, identification, and learning of complex systems
IEEJ J Ind Appl
System identification: a frequency domain approach
System identification: Theory for the user
A data-driven approach to robust control of multivariable systems by convex optimization
Automatica
Accurate FRF identification of LPV systems: nD-LPM with application to a medical X-ray system
IEEE Trans Control Syst Technol
Optimal input signals for parameter estimation in dynamic systems–survey and new results
IEEE Trans Automat Control
Optimal experiment design for open and closed-loop system identification
Commun Inf Syst
Constrained optimal input signal design for data-centric estimation methods
IEEE Trans Automat Control
Dynamic system identification: experiment design and data analysis
Cited by (8)
Data-driven robust iterative learning control of linear systems
2024, AutomaticaApplication-Oriented Input Spectrum Design in Closed-Loop Identification
2023, Applied Sciences (Switzerland)Statistical Characterization of Position-Dependent Behavior Using Frequency-Aware B-Spline
2023, IEEE International Conference on Intelligent Robots and SystemsIterative Robust Experiment Design for MIMO System Identification via the S-Lemma
2023, 2023 IEEE Conference on Control Technology and Applications, CCTA 2023
Nic Dirkx received the M.Sc. degree from the Eindhoven University of Technology, Eindhoven, The Netherlands, in 2011. Since 2012, he is a mechatronics design engineer at ASML, Veldhoven, The Netherlands. As of 2018, he is also pursuing the Ph.D. degree in the Control Systems Technology group within the department of Mechanical Engineering at TU/e. His research interests are in the field of system identification, precision motion control, and mechatronic systems.
Jeroen van de Wijdeven received the M.Sc. and Ph.D degrees in Mechanical Engineering from the Eindhoven University of Technology, The Netherlands, in 2004 and 2008, respectively. Since 2008 he is a mechatronics design engineer at ASML, Veldhoven, The Netherlands. His research interests include motion control and dynamics for high-precision mechatronic systems.
Tom Oomen received the M.Sc. degree (cum laude) and Ph.D. degree from the Eindhoven University of Technology, Eindhoven, The Netherlands, in 2005 and 2010, respectively. He held visiting positions at KTH, Stockholm, Sweden, and at The University of Newcastle, Australia. Presently, he is associate professor with the Department of Mechanical Engineering at the Eindhoven University of Technology. He is a recipient of the Corus Young Talent Graduation Award, the IFAC 2019 TC 4.2 Mechatronics Young Research Award, the 2015 IEEE Transactions on Control Systems Technology Outstanding Paper Award, the 2017 IFAC Mechatronics Best Paper Award, the 2019 IEEJ Journal of Industry Applications Best Paper Award, and recipient of a Veni and Vidi personal grant. He is Associate Editor of the IEEE Control Systems Letters (L-CSS), IFAC Mechatronics, and IEEE Transactions on Control Systems Technology. He is a member of the Eindhoven Young Academy of Engineering. His research interests are in the field of data-driven modeling, learning, and control, with applications in precision mechatronics.