DUDLEYKNOXLIBRARY NAVALPOSTGRADUATESCHOOL MONTEREY CA 93943-5101 DUDLEYKNOXLIBRARY NAVAL POSTGRADUATE SCHOOL MONTEREY CA 93943-5101 Approved for public release; distribution is unlimited. Design of Robust Suboptimal Controllers for a Generalized Quadratic Criterion by Kurtis Brett Miller Lieutenant, United States Navy B.S., Colorado State University, 1986 Submitted in partial fulfillment ofthe requirements for the degree of MASTER OF SCIENCE IN ELECTRICAL ENGINEERING from the NAVAL POSTGRADUATE SCHOOL June 1992 Computer Engineering 11 classified ORTTY CLASSIFICATION OFTHIS PAGE REPORT DOCUMENTATION PAGE Form Approved OMB No. 0704-0188 IEPORT SECURITY CLASSIFICATION lb. RESTRICTIVEMARKINGS Unclassified :ECURITY CLASSIFICATION AUTHORITY 3. DISTRIBUTION/AVAILABILITY OF REPORT Approved forpublic release; distribution is unlimited. DECLASSIFICATION/DOWNGRADING SCHEDULE ERFORMING ORGANIZATION REPORT NUMBER(S) 5. MONITORING ORGANIZATION REPORTNUMBER(S) vJAME OF PERFORMING ORGANIZATION 6b. OFFICE SYMBOL 7a. NAMEOFMONITORING ORGANIZATION (Ifapplicable) ^faval Postgraduate School 33 Naval Postgraduate School AMDoDnRtEeSrSey(C,ityC, SAtate,9a3nd94ZI3P-C5od0e)00 7b.MAoDnDtReESrSey(C,ityC, SAtate,9a3n9d4ZI3P-C5od0e0)0 SAME OF FUNDING/SPONSORING 8b. OFFICE SYMBOL 9. PROCUREMENTINSTRUMENTIDENTIFICATION NUMBER ORGANIZATION (Ifapplicable) ADDRESS (City, State, and ZIP Code) P10R.OSGORURACMEOFFUNPDRIONJGENCTUMBERS TASK WORKUNIT ELEMENTNO. NO. NO. ACCESSION NO. TDIETLSEI(IGncNludOe SFecuRritOyBClaUssiSfiTcatiSon)UBOPTHvIALCONTROLLERS ^F^O„R A, G^.TE-vN-rEr-Rr.AaLtIxrZzrE-rD^ QrvUTTAADnRnAATTIT/C"> :RTrERION - . PERSONAL AUTHOR(S) <.urtis B. Miller TYPE OF REPORT 13b. TTME COVERED 14. DATE OF REPORT (Year,Month,Day) 15. PAGE COUNT Master's Thesis FROM TO June 1992 66 DSiUPePvLEiMeEwNsTAeRxYprNeOTsAsTeIdONin this thesis are those ofthe author and do not reflect the offici.al. pol..icy orpo«s-Si«tJi/o«n, rofS he Department ofDefense or the U.S. Government. COSATI CODES 18. SUBJECT TERMS (Continue on reverse if necessary and identify by block number) FIELD GROUP SUB-GROUP Linear Quadratic Feedback Control System, RobustControl, Optimal Control, Suboptimal Control ABSStTaRAnCdTar(Cdonltiinnueeaornqruevaerdsreaitfineccersseagryulanadtoidrent(ifLyQbyR)blodckesnuimgbenrs) guarantee acertai_n level ofrobustness. However, Dptimizing a generalized quadratic criterion produces coupled state and input terms and there are no longer any guarantees ofgood robustness properties. This thesis identifies how this problem arises and then presents several suboptimal, but robustcontrollerdesign options which provide the control systems engineer with the ability to perform a trade-offbetween performance and robustness. The effectivness of these methods is investigated and the trade-offs between performance and robustness are evaluated usmg computer simulation ofa statically unstable fighteraircraft. DISTRIBUTION/AVAILABILITY OF ABSTRACT 21. ABSTRACT SECURITY CLASSIFICATION x] UNCLASSIPIED/UNIJOVnTED ] SAME AS RPT. j DTICUSERS Unclassified l NAMEOF RESPONSIBLEINDIVIDUAL 22b. TELEPHONE (Include Area Code) 22c. OFEFICCE/SCYwMBOL Won-Zon Chen _^__ (408) 646 - 2928 ) Form 1473, JUN 86 Previous editions are obsolete. SFCI1RTTY CLASSIFICATION OFTHIS PAGE S/N 0102-LF-014-6603 Unclassified ABSTRACT Standard linear quadratic regulator (LQR) designs guarantee a certain level of robustness. However, optimizing a generalized quadratic criterion produces coupled state and input terms and there are no longer any guarantees of good robustness properties. This thesis identifies how this problem arises and then presents several suboptimal, but robust controller design options which provide the control systems engineer with the ability to perform a trade-off between performance and robustness. The effectiveness of these methods is investigated and the trade-offs between performance and robustness are evaluated using computer simulation of a statically unstable fighter aircraft. ra TABLE OF CONTENTS INTRODUCTION l I. BACKGROUND LINEAR QUADRATIC THEORY 3 II. STANDARD LINEAR QUADRATIC CONTROL THEORY 3 A. ROBUSTNESS PROPERTIES OF LQ DESIGNS 5 B. THE GENERALIZED QUADRATIC CRITERION 8 C. ROBUSTNESS FOR A GENERALIZED QUADRATIC D. 12 CRITERION MODEL UNCERTAINTY AND ROBUSTNESS MEASURES III FOR MULTIVARIABLE SYSTEMS 15 MMO MODEL UNCERTAINTY FOR SYSTEMS 15 A. ROBUSTNESS MEASURES FOR MIMO SYSTEMS 17 B. 17 Principal Region x ' 1. 19 Minimum Singular Value 2. ROBUST SUBOPTIMAL DESIGN 22 IV. ROBUSTNESS DESIGN OPTIONS 22 A. 23 1. Option I: S = Q = Q +SS\ R = R+ 23 2. Option B: I R =pR 23 3. Option III: = p<l 24 4. Option IV: S pS, 5. Option V: R = pR, Q = Q+(P-DSR^S*, p> 1 24 6. Option VI: Q = pQ 25 Option VB: Q = Q+pSR_1S*, p>l 26 7. A NUMERICAL EXAMPLE AND SIMULATION 27 V. STATICALLY UNSTABLE AIRCRAFT MODEL 27 A. w 1 DUDLEYKNOXLIBRARY NAVALPOSTGRADUATESCHOOL MONTEREY CA 93943-5101 B. SIMULATIONS 33 Optimal System 34 1. 2. System with Q = Q+SS* R= R+ 36 , 3. System with R =pR 38 4. System with S = pS, p < 1 39 5. System with R =pR, Q = Q+(p-l)SR_1S*, p>l 42 6. System with Q =pQ 45 7. System with Q = Q +pSR._-!1«S.* p > 1 48 , C. FURTHER INVESTIGATION OF OPTION IV 51 CONCLUSION VI. 54 REFERENCES 56 INITIAL DISTRIBUTION LIST 57 LIST OF FIGURES Figure 2.1 Closed-Loop Linear System 4 Figure 2.2 Definitions ofRobustness Measures [After Ref. 3] 6 Figure 2.3 Robustness ofan LQ System [AfterRef. 3] 8 Figure 3.1 System with Unstructured Uncertainties 17 MEMO Figure 3.2 Principal Region and Stability Margins for systems [After Ref. 4] 18 Figure 5.1 Pitch Pointing, Constant Flight Path 28 Figure 5.2 DirectLifting, ConstantPitch 28 Figure 5.3 Definitions of State Variables a, 8, and q 29 Figure 5.4 Wind Gust Disturbance Inputs 33 Figure 5.5 Angle ofAttack 33 Figure 5.6 Minimum Singular Values of the Optimal System 34 Figure 5.7 Response of the Optimal System to a Rectangular Gust 35 Figure 5.8 Response ofthe Optimal System to a Triangular Gust 35 Figure 5.9 Minimum Singular Values Using Q = Q +SS* R = R+1 36 , Figure 5.10 Square Gust Response of Q = Q+SS*, R = R+I 37 Figure 5.11 Triangular Gust Response of Q =Q+SS*, R=R+I 37 Figure 5.12 Performance vs p for R= pR 38 Figure 5.13 Robustness vs pforR = pR 38 Figure 5.14 Performance vs p for S =pS, p < 1 39 Figure 5.15 Robustness vs p for S = pS, p<l 39 VI