0 Up votes0 Down votes

7 views6 pagesSystemIdentificationUsingLaguerreFunctions

Jun 02, 2013

© Attribution Non-Commercial (BY-NC)

DOC, PDF, TXT or read online from Scribd

SystemIdentificationUsingLaguerreFunctions

Attribution Non-Commercial (BY-NC)

7 views

SystemIdentificationUsingLaguerreFunctions

Attribution Non-Commercial (BY-NC)

- Neuromancer
- The E-Myth Revisited: Why Most Small Businesses Don't Work and
- How Not to Be Wrong: The Power of Mathematical Thinking
- Drive: The Surprising Truth About What Motivates Us
- Chaos: Making a New Science
- The Joy of x: A Guided Tour of Math, from One to Infinity
- How to Read a Person Like a Book
- Moonwalking with Einstein: The Art and Science of Remembering Everything
- The Wright Brothers
- The 6th Extinction
- The Housekeeper and the Professor: A Novel
- The Power of Discipline: 7 Ways it Can Change Your Life
- The 10X Rule: The Only Difference Between Success and Failure
- A Short History of Nearly Everything
- The Kiss Quotient: A Novel
- The End of Average: How We Succeed in a World That Values Sameness
- Made to Stick: Why Some Ideas Survive and Others Die
- Algorithms to Live By: The Computer Science of Human Decisions
- The Universe in a Nutshell

You are on page 1of 6

Olivier Department of Electrical and Computer Engineering Mercer University Macon, GA 31211 Olivier_pd@mercer.edu

Abstract A system identification technique based on Laguerre series expansions of filtered input/output data is briefly summarized. Some simple examples are presented that demonstrate the power of the technique. The examples verify that the procedure works well for lightly damped (i.e. flexible) systems, for MIMO systems as well as SISO systems, and that non positive real systems pose no problems. 1. Introduction There are many situations in which it is necessary to approximate the transfer function model of a physical system from input/output data. This problem arises in adaptive control, deconvolution, fault detection and many other areas. This paper describes a new way to find transfer function models based on input/output time domain data. Procedure Description. Transfer function models of systems are obtained from input/output data based on the Laguerre series expansions of filtered input/output data. The steps in the technique are as follows: 1) find the coefficients for the Laguerre series representation of the filtered input and output data, this can be done using either time domain or frequency domain data; 2) if time domain data is used, the time domain representations are Laplace transformed yielding frequency domain representations of the filtered input and output, U ( s ) , and Y ( s ) ; 3) the unknown transfer function matrix is assumed to have the form A( s ) B ( s ) 1 where A( s ) and B ( s ) are assumed to have Laguerre series expansions with unknown coefficients; 4) the products B ( s )Y ( s ) and A( s)U ( s) are expanded in terms of products of Laguerre functions; 5) the products of Laguerre functions are replaced by differences of Laguerre functions; 6) coefficients of like Laguerre functions are equated; and 7) solving these equations yield the coefficients of A( s ) and B ( s ) . Paper organization. The remainder of this paper is organized as follows: Section 2 summarizes the pertinent facts about Laguerre functions; Section 3 presents the examples; Section 4 provides the summary and discussion of the results

2. Laguerre Functions The Laguerre functions are an orthonormal set of functions that are complete in the function space L2 . The Laplace transform of any of these functions is a rational function of the Laplace variable s , that has all of its poles at the a point on the negative real axis. In the time domain, the Laguerre functions are polynomials multiplied by a decaying exponential. As such, the Laguerre functions can be used to approximate stable transfer functions, and/or reasonably behaved functions that decay to zero in the time domain. The basic facts about these functions are summarized below. Frequency domain description. The k th Laguerre function is

Lk ( s ) = 2 p ( s p ) k 1 ( s + p) k

pt / 2

k !( pt )n 1 2 n =1 ( n 1)!) ( k n + 1)!

k +1

Expansion equations. A reasonably well behaved function can be expanded in terms of the Laguerre functions as

f (t ) = a k Lk (t ) F ( s ) = a k Lk ( s )

k= 1 k= 1

where the coefficients a k can be obtained in either the time or frequency domain via standard inner product computations.

where the time domain inner product is the standard L2 inner product and the s domain inner product is the L2 inner product induced on the s domain by Parsevals theorem, i.e.

< f (t ), g (t ) >t = f (t ) g (t ) dt

0

< F ( s ), G ( s ) >s =

1 2

F ( s)G (s)ds

where k s n is the standard dirac delta. Product property. In addition to these standard properties of Laguerre functions, this paper relies on the following newer property [see Olivier, 1995] that states that a product of two s domain Laguerre functions is the weighted difference of two other s domain Laguerre functions.

Lk ( s ) Lm ( s ) = Lk +m 1 ( s ) Lk +m ( s ) 2p

3. Examples Three simple examples are considered. These examples are chosen to demonstrate that the procedure will work on examples with known solutions. The first example mimics some properties of flexible structures, the second example is a MIMO system, and the third example is of a non positive real system (a stumbling block for some other system ID procedures). Flexible structure example. NASAs Marshall Space Flight Center has a flexible structure testbed referred to as CASES. This structure was designed to exhibit many of the properties of flexible structures that cause problems to control engineers. The example considered here is motivated by the CASES structure in that the transfer function to be identified has damping (very low), and natural frequency (low) similar to the average damping and natural frequency of the CASES structure. Consider a system described by the transfer function H ( s ) = ( s + a ) /( s + a ) 2 + b 2 ) with a = .7084 and b = 7.6469 . The pole of the Laguerre functions was chosen to bp = a . The input was chosen to be a step input and the input and output data were prefiltered by a filter with transfer function F ( s ) = s /( s + p ) . The procedure outlined in the introduction is implemented to yield the following: Step 1. Expand filtered input: u (t ) = u1 L1 (t ) + u 2 L2 (t ) + ... = (1 /( 2 p ) 5 ) L1 (t ) Expand the filtered output: y (t ) = y1 L1 (t ) + y 2 L2 (t ) + ... + y10 L10 (t ) Step 2. Laplace transform these time domain representations: U ( s ) = (1 /( 2 P ) 5 ) L1 ( s )

Y ( s ) = y1 L1 ( s ) + y 2 L2 ( s ) + ... + y10 L10 ( s )

Step 3. Assume Laguerre expansions for the numerator and denominator A( s ) and

B( s) A( s ) = a1 L1 ( s ) + a 2 L2 ( s ) + a 3 L3 ( s ) B( s ) = b1 L1 ( s ) + b2 L2 ( s ) + b3 L3 ( s )

These expansions were chosen to be 3rd order to get 2nd order approximations. Step 4. Expand the products A( s)U ( s ) and B ( s)Y ( s ) : B ( s )Y ( s ) = [ b1 L1 ( s ) + b2 L2 ( s ) + b3 L3 ( s )] [ y1 L1 ( s ) + y 2 L2 ( s ) + ... + y10 L10 ( s )]

= b1 y1 L1 L1 + b1 y 2 L1 L2 + b1 y 3 L1 L3 +... + b3 y1 L3 L1 + b3 y 2 L3 L2 + b3 b3 L3 L3

= (1 / 2 p ) 5 )[ a1 L1 ( s ) L1 ( s ) + a 2 L2 ( s ) L1 ( s ) + a 3 L3 ( s ) L1 ( s )]

A( s )U ( s ) = (1 /( 2 p ) 5 )[ a1 L1 ( s ) + a 2 L 2 ( s ) + a 3 L3 ( s )] L1 ( s )

Step 5. Replace products of Laguerre functions with differences of Laguerre functions using the product property.

B ( s )Y ( s ) = [b1 y1 ( L1 L2 ) + b1 y 2 ( L2 L3 ) + b1 y 3 ( L3 L4 ) + ...

+b3 y1 ( L3 L4 ) + b3 y2 ( L4 L5 ) + b3 y2 ( L4 L5 ) + b3 y3 ( L5 L6 )]/ 2 p A( s)U ( s ) = [a1 ( L1 L2 ) + a2 ( L2 L3 ) + a3 ( L3 L4 )] / 2 p = [ a1 L1 + (a2 a1 ) L2 + (a3 a2 ) L3 + a3 L4 ] / 2 p Step 6. Equate coefficients of like Laguerre functions. This produces a set of matrix equations of the form [Y U ] x = 0 where the first column of Y is Y (i,1) = y1 Y (i,1) = y 2 y1 , if j < i The other columns are shifted down so that the non square matrix Y is lower triangular. The matrix U has the same structure. The unknown vector x is composed of x = [b, a , ], . Step 7. Solve the equations. The set of equations have nontrivial solution only if they are underdetermined. To obtain a non singular set of equations, bring all terms involving b3 (or some other b) to the right, and arbitrarily set b3 = 1 . This yields [ Y (:,1: 2) U ] z = Y (:,3) where z is x with b3 removed. This is the set of equations that is solved. The technique identified the transfer function to 4 significant figures yielding H a ( s) = (1.0004 s + .7128) /( s 2 + 1.4141s + 58.9963) 7.0764e5 MIMO example. Multi input and/or Multi output systems can be handled just as easily. Consider a system described by the MIMO transfer function

2 s + 3 s + 2 4 s + 3 s + 3 H (s) = ( s + 1)( s + 2)

The expansions apply without modification, except that the coefficients are matrices, and that care must be taken to keep the order of matrix products correct. Step inputs are applied one at a time to identify each column separately.

The technique again identified the original transfer function quite accurately, yielding

1.0000( s + 2) 2.000s + 3.0002 4.0001s + 3.000 1.0000s + 3.0000 1.0000 s 2 + 3.0000 s + 2.0000

H a (s) =

7e 6 E a ( s) = 2e 5 3e 6 2e 6

NPR example. Non positive real transfer functions cause problems for many system identification procedures. This example presents the results of identification of a non positive real system. Consider the transfer function H ( s ) 1/(( s + a ) 2 + b 2 ) with the same a, and b as in the first example. The technique produced

H a ( s ) = (.0003s + 1.0044) /( s 2 + 1.4293s + 59.0671) 4.9e 5

4. Summary/Discussion This paper has outlined a system identification procedure based on Laguerre series expansions of system input and output. It has demonstrated that the technique works well for simple SISO, MIMO and NPR examples. Some particular points to be made include: a) This approach can be viewed as a numerical Laplace transform b) All unknown parameters enter linearly, i.e. no nonlinear solution techniques are needed. Future work needs to include input and output noise. Future applications appear to include: adaptive control of flexible structures; fault detection and isolation in analog circuits; crack detection in aircraft. 5. Acknoledgements This research was supported in part by a grant from the National Aeronautics and Space Administration through grant no. NAG8-1301. The author greatly appreciates this support. 6. References 1. Olivier, P.D., On-line system identification using Lagueer Series, IEEE Proceedings D, Control Theory and Applications, 1994, Vol. 141, (4), pp. 249-251. 2. Olivier, P.D., Approximate System Inverses, Electronics Letters, 1995, Vol. 31, No. 23, pp. 2050-2051.

- Introduction to MatlabUploaded byXxbugmenotxX
- EC409 Control Systems (CareerYuga)Uploaded byjinto0007
- Best Int ControlUploaded byhinsermu
- RoofUploaded byshingkeong
- Senesky Tsao Modelling Control Synchronous MachineUploaded byiMiklae
- 1506.06218Uploaded byhuevonomar05
- Ieee Ccac 2017 Paper 35Uploaded byByron Paul Cajamarca
- Network AssignmentUploaded byKeep Greenery
- Control SystemUploaded byZahid Mehmood
- allugtopicsUploaded byMohsen Nasseri
- PM11 Airturn Temp T_200_201 ConfigUploaded byFrebu Trilangga
- Zaccarian_talk2Uploaded byEric Flores
- hw01Uploaded byAlexander Bennett
- Auotmation TechnologyUploaded byguptamg87
- The International Journal of Applied Control, Electrical and Electronics Engineering (IJACEEE)Uploaded byamsjournal
- PPT_FW3Uploaded byAzito Jum
- COBACANAUploaded byLu Mateo
- pidUploaded byrobinrastogigcet
- Induction Motor Speed COntrol Base on Model ReferenceUploaded byLuis Angeles
- Pawan Model Optimal Control of the Four Tank SystemUploaded byPawan Kumar Maurya
- 04118389Uploaded bymipanduro7224
- Mini ProjectsUploaded byAnonymous bY5z1z6a
- Global Adaptative Output Feedback Control of Induction Motor WithUploaded bysefibot
- Lead Compensator-Time DomainUploaded byberkut samaka
- ie0602815Uploaded bydrgry
- FloBoss104 OverviewUploaded byangeljos
- LFC PID 2Uploaded bySouvik Ganguli
- 14.a Simplified Control Technique for a Dual Unified Power Quality Conditioner (1)Uploaded byImran
- Circuito RCUploaded byJones Castro
- ME_Chemical 2013 Course.pdfUploaded byShivam Markad

- Key SpaceX RecommendationsUploaded byparikshat7
- Quadrone i-sight Manual BookUploaded byRaul Gomez FUTC
- Trent 700_tcm92-5733Uploaded byarabysadek
- BGTL.pdfUploaded bydebert
- 52056796 Rev 01Uploaded byLeonard
- NASA Space Shuttle STS-99 Press KitUploaded byOrion2015
- IITBadfUploaded bytinkudhull
- Impact of JetUploaded byOmar Faruqi
- 9 18995 01 VESDA VLI Installation Sheet LoresUploaded byEduardo Parrudo
- Fretting-initiated fatigue in large bore engines connecting rodsUploaded bysunita45
- INTRODUCTION (2) (Autosaved).docxUploaded bykwanele
- Gas TurbinesUploaded bymurari
- DEF_STAN_00-970_P1_ISSUE-5_SEC-3Uploaded byRamona Neculai
- The Study on the Performance of the Gas Turbine for Power GeUploaded byGal
- HIRM Design SpecUploaded bycmpmarinho
- 02e_Armour Layer Design Meynord US ArmyUploaded byAlejandro Urquieta Quiroga
- Weldwell Catalogue 2015Uploaded byjaganelr
- Formula Sheet Aircraft TutorialsUploaded byPythonraptor
- NASA Facts B-200 King Air Research Support AircraftUploaded byBob Andrepont
- Probability of Detection CurvesUploaded byDav89
- Ada 584825Uploaded byAbhishekKumar
- Ch1- Jigs & Fixtures.pptUploaded byttetslm
- case_c3.6Uploaded byMarco Aurélio Leonel Matunaga
- Vtu e Notes Mpiii 11Uploaded byHarish H
- Aviation Emergency Responsen Aircraft Guidebook CompressedUploaded bybroncatio
- Hkd GliderUploaded bydiv850
- Principles of Flight Practice ExamUploaded bythowmas
- 117352033 FutureAirlinerStudy Raymer PrepublicationCopy Rev9 2011Uploaded bynd1977
- Next War Korea Game Specific RulesUploaded byMitchell Land
- Aviation TheoryUploaded byFatima Postasa

## Much more than documents.

Discover everything Scribd has to offer, including books and audiobooks from major publishers.

Cancel anytime.