Bridging the theory-practice gap through problem reformulation: a motion control case study Zhiqiang Gao, Ph.D. Center for Advanced Control Technologies Cleveland State University June 24, 2004 www.cact.csuohio.edu 1 Outline • Introduction • The Theory-Practice Gap • An Experimental Science Approach to Control Research • Problem Reformulation • Conclusions 2 Center for Advanced Control Technologies From Applied Research to Advanced Technologies www.cact.csuohio.edu 3 CACT Mission • Define, Articulate, Formulate Fundamental Industrial Control Problems • Solutions and Cutting Edge Technologies • Performance and Applicability • Synergy in Research and Practice 4 Center for Advanced Control Technologies FACULTY: Dr. Zhiqiang Gao, Director of CACT, ACRL and AERL Dr. Daniel Simon, Director of the Embedded Systems Laboratory in Electrical Engineering. Dr. Paul Lin, Director of the 3D Optical Measurement Laboratory in Mechanical Engineering. Dr. Yongjian Fu, Software Engineering Dr. Sally Shao, Mathematics Prof. Jack Zeller, Engineering Technology, 40 yrs+ experience, P.E. 5 Center for Advanced Control Technologies Doctoral Candidate Researchers: Frank Goforth Robert Miklosovic Zhan Ping Wankun Zhou Aaron Radke Chunming Yang Qing Zheng Sri Kiran Kosanam Masters Candidate Researchers: Eric Dittmar Bharath Endurthi Hrishikesh Godbole Sai Kiran Gumma Qing Guo Ivan Jurcic Srujan Kusumba Mike Gray Xiaolong Li Ramgopal Mushini Nuha Nuwash Tong Ren Bhavinkumar Shah Chirayu Shah Madhura Shaligram 6 Past Projects • • • • • • • • • • Temperature Regulation Intelligent CPAP/BiPAP Motion Indexing Truck Anti-lock Brake System Web Tension Regulation Turbine Engine Diagnostic Computer Hard Disk Drive Stepper Motor Field Control 3D Vision Tire Measurement Digitally Controlled Power Converter 7 Sponsors • • • • • • • • • • NASA AlliedSignal Automotive Invacare Co. Energizer Rockwell Automation Kollmorgan ControlSoft Black and Decker Nordson Co. CAMP 8 NASA Intelligent PMAD Project 9 Networked Power Converters 10 Case Study: Web Tension Regulation 11 Case Study: Truck Anti-lock Brake System 12 Case Study: Computer Hard Disk Drive 13 We build it, test it, and make it work. We apply our research using our partner’s products. 15 We get results: Wavelet control for robust machines. 16 We get results: Advanced motor field control reduces cost. 17 We get results: Model Independent control design & tuning. 18 We write the software. 19 We have staff from the “School of Hard Knocks”. 20 It All Comes Down To Mathematics • Level of abstraction • Clarity in thinking • Theory and guidance 21 Theory vs. Practice A Historical Perspective 22 The Classical Control Era Control Practice Control Research Mathematics Control Theory 23 The Modern Control Era Control Practice Control Research Mathematics Control Theory 24 The transition did go quietly 25 4/1964 IEEE Trans. Automatic Control Editorial “In recent years, there has been considerable discussion about the gap which appears to exist between control theory and its application…” “It appears that the problem of the gap is a control problem in itself; it must be properly identified and optimized through proper action…” AACC Theory and Applications Committee meeting, 3/24/1964 “Bridging the Gap Between Theory and Practice” 26 4/1965 IEEE Trans. Automatic Control Guest Editorial by Harold Chestnut Proposed Solutions: • Company sponsored Education • Component Study by Universities • Promotion of Economic Incentives • Publication Policies • Definition of characteristics of systems and subsystems 27 8/1967 IEEE Trans. Automatic Control Editorial by J.C. Lozier On panel discussions: “These panels, staffed with leading theoreticians, have automatically assumed that theory is ahead of practice, and they conclude that the solution lies in reeducating the designers. The establishment has spoken.” Suggestion in bridging the gap: stimulate – papers on general practice – papers on advanced engineering practice – a more hospital atmosphere where results are as important as methods 28 2/1968 IEEE Trans. Automatic Control Announcement by John. B. Lewis • Special two day conference preceding JACC meeting • Responding to 65 and 67 editorials • Each session is “a complete case history necessarily brings together theory and practice” • “Demonstration of what can be achieved by applying control theory to major problems” 29 12/1982 IEEE Trans. Automatic Control Editorial by Y.C. Ho • “Control” as experimental science (the 3rd dimension w.r.t. the gap) • Experiment vs. Application (detective vs. craftsman) • The “observation-conjectureexperiment-theory-validation” paradigm • Carried out by BOTH theorists and experimentalists 30 The debate continues • “On Control Theory and Practice”, G. John, AC, June 1970 • “Editorial: Some Thoughts on Research”, J. Mereditch, AC, Feb. 1980 • “Editorial: Theory and application: A common ground?” M. Sain, June 1980. • “An Industrial Point of View on Control Teaching and Theory”, E. H. Bristol, CSM, Feb. 1986 • Special Issue on Theory and Practice Gap, CSM December 1999. • Theory vs. Practice Forum, ACC 2004 31 The ISA Initiative • ISA (Instrument, systems, and automation) is the largest organization of instrument and control engineers in the world • ISA is organizing a Theory vs. Practice Forum at ACC2004 (by Z. Gao and R. Rhinehart) 32 Reflection on Control Research What and Why? 33 What is controls? r e u y Controls: An instrument or a setPlant of Controller Reference Gas Pedal Vehicle Error Speedinstruments used to operate, Position regulate,Speed or guide a machine or vehicle Sensor -the American Heritage Dictionary Is it a branch of engineering, science, or mathematics? 34 Control Engineering? Engineering: The application of scientific principles to practical ends as the design, construction, and operation of efficient and economical structures, equipment, and systems. -the American Heritage Dictionary If control is a branch of engineering, what are the scientific principles behind it? 35 Control Science? Science: The observation, identification, description, experimental investigation, and theoretical explanation of natural phenomena. -the American Heritage Dictionary 36 The Theory-Practice Divide • Practitioners practice, improvise, experiment (experience counts in industry) • Theorists theorize Modern Control Theory is often viewed as a branch of Applied Mathematics • Why the divide? 37 Experimental Controls Research Discover vs. Apply 38 Experiment Discover Theorize 39 • Observation: 95% of controllers used in Industry is PID u K p e K I edt K D e • Conjectures: • Error based design must have merits; • Solution to robust control is outside the realm of modern control theory; • Better controllers can be found experimentally 40 Experiment #1: A design not strictly based on the math model y f (t , y, y, w) bu f (t , y, y, w) f (t ) u ( f (t ) u0 ) / b y u0 41 A unique disturbance estimator Augmented plant in state space: x1 y, x2 y, x3 f (t , y, y, w) y f (t , y, y, w) bu x1 x2 x x bu, 3 2 x3 f y x 1 Extended State Observer z1 z2 1 g1 ( z1 y ) z2 z3 2 g 2 ( z1 y ) bu z g ( z y) 3 3 1 3 z1 x1 z2 x2 z3 x3 42 Active disturbance compensation x1 x2 x2 f bu y x 1 u (u0 z3 ) / b z3 f f (t ) or f (t , x1 , x2 , w)? x1 x2 x2 u0 y x 1 43 Active Disturbance Rejection Control w(t) v2(t) r(t) u0(t) +_ Profile Generator +_ Nonlinear PD u(t) y(t) +_ Plant 1/b0 b0 v1(t) z3(t) z2(t) Extended State Observer (ESO) z1(t) 44 A Breakthrough in Motion Control y f (t , y, y, w) bu (t ) transient profile and output 2 bandwidth: 4 rad/sec bandwidth: 20 rad/sec transient profile 1 0 position 2 0 1 2 3 error 4 5 6 0 1 2 control3signal 4 5 6 0 1 2 3 time second 4 5 6 1 y z1 1 0 0.5 0 1 2 3 velocity 4 5 6 2 dy/dt z2 1 0 0 -1 2 0 1 2 3 4 disturbance and unknown dyanmics 5 6 50 f z3 1 0 0 -1 -50 0 1 2 3 time second 4 5 6 45 Hardware Test: torque disturbance Torque Disturbance Rejection Rev. 1.5 Position ADRC 1 PID 0.5 0 0 2 4 6 8 Rev. 0.1 10 12 10 12 PID Position error 0 ADRC -0.1 0 2 4 6 8 Volts 5 ADRC Control Command 0 PID -5 0 2 4 6 8 10 12 46 Performance of the disturbance observer Total disturbance and its estimation 30 20 a(t) f(t) z3(t) 10 0 -10 -20 -30 0 1 2 3 4 5 Time (sec.) 47 Extension to Higher Order MIMO Plants , Y ( n1) , d (t )) U , Y Rl ,U Rl , d R m Y ( n) F (Y , Y , Model of F(.) in the state space→ in the time domain: W (t ) F (Y , Y , , Y ( n 1) , d (t )) U W (t ) V Y (n) V How to reconstruct the extended state W (t )F (Y (t ), Y (t ), From U and Y ? , Y ( n1) (t ), d (t )) 48 Y ( n) F (Y , Y , , Y ( n1) , d (t )) U W (t ) F (Y , Y , Y (n) W (t ) U U W (t ) U 0 , Y ( n1) , d (t )) Extended state Dynamic linearization and decoupling Y (n) U 0 z1i z2i 01 g1 ( z1i yi ) z2i z3 02 g 2 ( z1i yi ) z z ni n 1,i 0 n g n ( z1i yi ) ui zn 1,i 0 n 1 g n 1 ( z1i yi ) Extended state observer (ESO) Z1 Y Z2 Y Z n Y ( n 1) Z n 1 W (t ) U Z n 1 U 0 49 Successful Applications • • • • • • • Motion Control (All manufacturing Industries) Web Tension Regulation (paper, steel, printing..) Machine Tools Power Electronics (Motor, Converters …) Aircraft Control (MIMO) Process Control (with long transport delay) Active Magnetic Bearing 50 New Control Technologies “Discovered” • • • • • Nonlinear PID Discrete Time Optimal Control Active Disturbance Rejection Single Parameter Auto-Tuning Wavelet Controller/Filter 51 A Paradigm Shift • Gao, Huang, Han, CDC2001 • Existing Paradigm: Model Based Design • New Paradigm: Error (e=r-y) Based • Same Objective: Desired error behavior 52 The Paradox of the Robust Control Problem Making the performance of the modeldependent control design independent of the model 53 GÖdel’s Incompleteness Theorem “Within any formal system of axioms, such as present day mathematics, questions always persist that can neither be proved or disproved on the basis of the axioms that define the system.” --paraphrased by S. Hawking 54 Is the solution to the robust control problem outside the existing control theory? 55 Problem Reformulation reconnect theory to practice 56 Reconnect Control Practice Control Research Mathematics Control Theory 57 Components of Problem Definition • Assumptions on the plant: – What is the minimum info needed for design? – What info is available in practice? • Design Objectives: – Absolute requirements – Criteria of optimality (judgment for comparison) • Design Constraints: – Actuator/sensor/digital controller – Hard and soft constraints 58 A Motion Control Case Study y f (t , y, y, w) bu f (t , y, y, w) : continuous and differentiable f (t , y, y, w) F | f (t , y, y, w) | k1,| f (t , y, y, w) | k2 b, k1, and k2 are given 59 A Common Design Objective Make y follow a reference signal, v, within a specified accuracy: |v - y| < g(v,t), g(v,t) > 0 is given a special case: g(v,t) is a constant V {v :| v | v 0 ,| v | v1,| v | v 2 } 60 Common Design Constraints Hard constraints | u | umax ,| u | umax Soft constraints u is “smooth” t2 | u (t ) | dt is “small” t1 61 Controller C(p) • A dynamic system represented in s.s. as z p( z, u, v, v, y) u q( z, v, v, y) • p is the parameter vector to be selected (tuned) 62 Problem Formulation f (t , y, y, w) F v V • Does a controller C(p) exist that meets the design objective subject to the constraints? • If so, how to find it? • If there is more than one solutions, what is the optimal solution in practical sense? • How to find such an optimal solution? 63 Where are we? Observations Conjectures Experiments • Theory? – formulate the problem • Validation? 64 Build A New Research Infrastructure • Practitioners/Researchers/Mathematicians • Discover (both practitioners and theoreticians) • Theorize – Prove stability and convergence – Generalized – Establish a new kind of theory • Validate – Verify the new theory against other problems – Define the range of applicability 65 Conclusions • Think out of box: controls as an experimental science • Experiments lead to new methods • From problems to methods to theory 66 A Nonlinear PID Application NPID CONTROLLER Transient Profile PWM Kp Ki 1 s Kd s ( s 1)2 DC-DC Converter Output Voltage Signal Conditioning 67 Load Disturbance Rejection Load decreasing(36->3A) Load increasing(3->36A) PI PI NPID NPID 68 PI NPID Output Voltage 1.0V/div ADRC TOADRC Time: 2.0ms.div 69 Discrete Time Optimal Control Law u fst (( x1 , x2 , r , h) x ( k 1) Ax ( k ) Bu ( k ) d rh; d 0 hd | u ( k ) | r y x1 hx2 x 1 x ,A x 0 x (0) x0 a0 d 2 8r | y | 1 2 0 xf 0 h 0 ,B h 1 a0 d sign( y ), | y | d 0 x2 a 2 x2 y / h, | y | d 0 r sign(a ), | a | d fst a | a | d r d , 70 Comparison of switching curves 71 72 73 Other Nonlinear Feedbacks (not based on Lyapunov methods) • Explore the use of nonlinear mechanisms – Nonlinear feedback – Nonlinear differentiation – Nonlinear PID – Discrete Time Optimal Control 74 Nonlinear PID u K p g p (e) K I gi (e)dt K D g d (e) • Nonlinear “proportional” term gp(e) – Small error, large gain – Reduce the role of integrator • Nonlinear integral control – Reduce phase lag – Maintain zero s.s. error and good disturbance rejection • Nonlinear differentiator – Noise immunity 75 Non-smooth feedback y w u, w 0, u 10 e sign(e), e r y a 26 76 Non-smooth feedback y w u, w 2 sin(10t ), u 10 e sign(e) a 27 77 Further gain experimentation u u=e u = fal(e) -d d 1 e | e |a sign(e), | e | d , fal (e) d 0 1 a | e | d , e/d , 78 A special case of NPD • Time Optimal Control (TOC) of a double integral plant • Solution obtained in the 60s for continuous plants • Chattering problem • Recent solution (DTOC) for discrete plant fundamentally resolved the chattering • Used as a controller or a differentiator 79