Data-Driven Modeling, Filtering and Control: Methods and applications

Description

The scientific research in many engineering fields has been shifting from traditional first-principle-based to data-driven or evidence-based theories. The latter methods may enable better system design, based on more accurate and verifiable information. In the era of big data, IoT and cyber-physical systems, this subject is of growing importance, as data-driven approaches are key enablers to solve problems that could not be addressed by standard approaches. This book presents a number of innovative data-driven methodologies, complemented by significant application examples, to show the potential offered by the most recent advances in the field. Applicable across a range of disciplines, the topics discussed here will be of interest to scientists, engineers and students in automatic control and learning systems, automotive and aerospace engineering, electrical engineering and signal processing. Cover Contents 1 Introduction 1.1 Introduction 1.2 State-of-the-art 1.3 Goals and structure of the book References Part I Data-driven modeling 2 A kernel-based approach to supervised nonparametricidentification ofWiener systems 2.1 Introduction and motivation 2.2 Preliminaries 2.2.1 Notation and definitions 2.2.2 Solving polynomial optimization problems via convex optimization 2.2.3 Exploiting sparsity in polynomial optimization 2.3 Problem statement 2.4 Maximum margin Hankel classifiers 2.4.1 Further computational complexity reduction 2.4.2 Exploiting sparsity 2.5 Examples 2.5.1 Synthetic data 2.5.1.1 Performance ofW as a classifier 2.5.1.2 Advantages of using negative sequences during training 2.5.2 Application: activity recognition from video data 2.6 Conclusions Acknowledgments References 3 Identification of a quasi-LPV model for wing-flutter analysis using machine-learning techniques 3.1 Introduction 3.2 LPV state-space model parameterization 3.3 Model estimation 3.3.1 Parameter reconstruction 3.4 Ensemble estimation approach 3.5 Wing-flutter model identification 3.6 Concluding remarks References 4 Experimental modeling of a web-winding machine: LPV approaches 4.1 Introduction 4.2 Sparse set membership identification of state-space LPV systems 4.3 Interpolated identification of state-space LPV systems 4.4 Web-winding system identification 4.4.1 The web-winding system 4.4.2 Experiment description 4.4.3 Sparse set membership LPV model 4.4.4 Interpolated LPV model 4.4.5 Model validation and results 4.5 Conclusion References 5 In situ identification of electrochemical impedance spectra for Li-ion batteries 5.1 Introduction 5.1.1 Motivation: understanding battery dynamics 5.1.2 Traditional methods for measuring EIS 5.1.3 Related work 5.1.4 Outline of approach 5.2 Method 5.2.1 Data collection 5.2.2 Identification 5.2.3 Frequency response and uncertainty estimation 5.2.4 Combined frequency response estimate 5.2.5 Review of frequency identification method 5.3 Example experimental results 5.3.1 Experimental conditions for PRBS perturbation 5.3.2 Experimental conditions for sinusoidal perturbation 5.3.3 Results Acknowledgments References Part II Data-driven filtering and control 6 Dynamic measurement 6.1 Introduction 6.1.1 Literature review 6.2 Problem setup 6.3 Model-based vs data-driven approaches 6.4 Maximum-likelihood data-driven estimation method 6.5 Examples 6.5.1 Methods and evaluation criterion 6.5.2 Example of temperature measurement 6.5.3 Example of mass measurement 6.5.4 Results 6.6 Conclusions and discussion Acknowledgments References 7 Multivariable iterative learning control: analysis and designs for engineering applications 7.1 Introduction 7.1.1 ILC for complex engineering applications 7.1.2 Design requirements for high-precision applications 7.1.3 Robust multivariable ILC design: the importance of (under) modeling (R1 – R2) 7.1.4 Model-free iterative learning (R2) 7.1.5 ILC for varying tasks (R3) 7.1.6 Contributions 7.1.7 Notation 7.2 System description and problem formulation 7.2.1 ILC framework 7.2.2 Convergence and performance 7.2.3 Design conditions for convergence and performance 7.2.4 Modeling considerations 7.3 ILC design—the SISO case 7.3.1 Manual design in the frequency domain 7.3.1.1 Design of L by inverting Ĵ 7.3.1.2 Design of Q based on FRF measurements 7.3.2 Design of learning filter: SISO inversion techniques 7.3.2.1 Approximate inversion 7.3.2.2 H∞-optimal synthesis with preview 7.3.2.3 Stable inversion 7.3.3 Toward MIMO ILC design: naive SISO design for MIMO systems 7.4 ILC Design—the MIMO case 7.4.1 Interaction analysis 7.4.2 Decoupling transformations 7.4.3 Robust multi-loop SISO design 7.4.4 Robust decentralized MIMO design 7.4.5 Centralized MIMO design 7.4.5.1 Design of learning filter: MIMO inversion techniques 7.5 Iterative inversion-based control: avoiding the need for parametric models 7.5.1 System description and procedure 7.5.2 Convergence analysis, modeling requirements and design 7.6 ILC with basis functions: enhancing flexibility to varying tasks 7.6.1 Flexibility in ILC—case study on a flatbed printer 7.6.2 Basis functions in ILC 7.6.3 Projection-based MIMO ILC with basis functions: frequency-domain design 7.6.3.1 Basis functions for MIMO ILC 7.6.3.2 Projection step 7.7 Conclusion and ongoing work Acknowledgments References 8 Algorithms for data-driven H∞-norm estimation 8.1 Motivation and problem formulation 8.1.1 Problem formulation 8.2 Power iterations 8.2.1 Power iterations in linear algebra 8.2.2 Power iterations for linear dynamical systems 8.2.3 An example 8.3 Multi-armed bandits 8.3.1 Stochastic multi-arm bandits in a nutshell 8.3.2 H∞-norm estimation as an MAB problem 8.3.3 Regret lower bounds and optimal algorithms 8.3.3.1 Regret lower bound in ΠSF 8.3.3.2 Regret lower bound in 8.3.4 The weighted Thompson sampling (WTS) algorithm 8.3.5 An illustrative example 8.4 Extensions to nonlinear systems 8.4.1 de Bruijn graphs and prime cycles 8.4.2 Finding the optimal stationary sequence 8.5 Discussion and extensions References 9 A comparative study of VRFT and set-membership data-driven controller design techniques: active suspension tuning case 9.1 Introduction 9.2 Problem statement 9.3 Controller tuning from data 9.3.1 Set-membership approach 9.3.2 Tuning via VRFT 9.4 Active suspension tuning case study 9.4.1 Controller tuning problem 9.4.2 Monte Carlo experiment 9.4.3 Process disturbance experiment 9.5 Conclusions Acknowledgment References 10 Relative accuracy of two methods for approximating observed Fisher information 10.1 Introduction 10.2 Background 10.2.1 The Central Limit Theorem 10.2.1.1 Lindeberg–Lévy CLT 10.2.1.2 Lyapunov CLT 10.2.1.3 Lindeberg CLT 10.2.2 Taylor expansion (Taylor series) 10.3 Theoretical analysis Scenario A. Independent and identically distributed (i.i.d.) samples Scenario B. Independent and nonidentically distributed (i.n.i.d.) samples 10.4 Numerical studies 10.5 Conclusions and future work 10.5.1 Conclusion 10.5.2 Future work AppendixA References 11 A hierarchical approach to data-driven LPV control design of constrained systems 11.1 Introduction 11.2 Related works 11.3 Problem formulation 11.4 A hierarchical approach 11.5 Data-driven inner controller design 11.5.1 Inversion of the reference model 11.5.2 Data-driven controller design 11.5.3 Dual problem 11.6 Outer controller design 11.7 Case study: servo-positioning system 11.7.1 System description 11.7.2 Desired inner closed-loop behavior 11.7.3 Inner controller design 11.7.4 Achieved inner closed-loop behavior 11.7.5 Outer controller design 11.8 Conclusions References 12 Set membership fault detection for nonlinear dynamic systems 12.1 Introduction 12.2 Nonlinear set membership fault detection 12.2.1 Problem formulation 12.3 Nonlinear set membership identification: global approach 12.3.1 Interval estimates 12.4 Nonlinear set membership identification: local approach 12.4.1 Interval estimates 12.4.1.1 Noise bounded in l2 norm 12.4.1.2 Noise bounded in l∞ norm 12.4.2 Local approach—identification algorithms 12.5 Nonlinear set membership identification: quasi-local approach 12.5.1 Interval estimates 12.6 Parameter estimation and adaptive set membership model 12.6.1 Parameter estimation 12.6.2 Adaptive set membership model 12.7 Summary of set membership fault-detection procedure 12.8 Example: fault detection for a drone actuator 12.8.1 Experimental setup 12.8.2 Nonlinear set membership fault detection 12.8.2.1 Global approach 12.8.2.2 Quasi-local approach 12.8.2.3 Local approach 12.9 Conclusions References 13 Robust data-driven control of systems with nonlinear distortions 13.1 Introduction 13.2 Preliminaries 13.2.1 Class of nonlinearities 13.2.2 Class of controllers 13.3 Frequency-domain identification 13.3.1 Stable plant 13.3.2 Unstable plant 13.3.3 Uncertainty filters for coprime factorisation 13.4 Robust controller design 13.4.1 Control performance 13.4.2 Convex formulation for robust performance 13.4.3 Controller design by convex optimisation 13.5 Case study 13.5.1 System description 13.5.2 Identification experiments 13.5.3 Performance specification 13.5.4 Experimental results 13.6 Conclusion References Index Back Cover