[m(^^MU^^ o 2 < < z NATIONAL ADVISORY COMMITTEE FOR AERONAUTICS TECHNICAL MEMORANDUM 1406 ON THE USE OF THE HARMONIC LINEARIZATION METHOD IN THE AUTOMATIC CONTROL THEORY By E. P. Popov Translation **K voprosu o primenenii metoda harmonicheskoi linearizatsii v teorii regulirovaniya." Doklady Akademii Nauk (SSSR), vol. 106, no. 2, 1956. UNfVERSfTY OF FLORIDA DOCUMENTS DEPARTMENT 1 20 MARSTON SCIENCE LIBRARY RO. BOX 117011 GAINESVILLE, FL 32611-701ll!JS.i\ Washington January 1957 NACA TM 1406 NATIONAL ADVISORY COMMITTEE FOR AERONAUTICS TECHNICAL MEMORANDUM 1406 ON THE USE OF THE HARMONIC LINEARIZATION METHOD IN THE AUTOMATIC CONTROL THEORY* By E. P. Popov The method of harmonic linearization (harmonic balance), first proposed by N. M. Krylov and N. N. Bogolyubov (ref . l) for the approxi- mate investigation of nonlinear vibrations, has been developed and re- eived wide practical application to problems in the theory of automatic control (refs. 3 to 6) . Recently, some doubt has been expressed on the legitimacy of application of the method to these problems, and assertions were made on the absence in them of a small parameter of any kind. Never- theless, the method gives practical, acceptable results and is a simple and powerful means in engineering computatations . Hence, the importance of questions arises as to its justification. The underlying principle of the method is the replacement of the given nonlinear equation by a linear equation. In establishing the method, a small parameter is considered whose presence makes it possible to speak, with some degree of approxi- mation, of the solution of this new equation to the solution of the given nonlinear equation. In an article by the author (ref. 7), certain con- siderations were given on the presence of the small parameter, but this question has not as yet received a final answer. In the present report, a somewhat different approach to the problem is applied that permits: (a) establishing, in the clearest manner, the form of the presence of the small parameter in nonlinear problems of control theory, solvable by the method of harmonic linearization; (b) connecting it with previous intuitive physical concepts (with the "filter property") and extending the class of problems possessing this property; and (c) discussing various generalizations of the method. The free motion (transition process and autovibration) for a very wide class of nonlinear systems of automatic control (ref. 7) are described by differential equations of the form Q(p)x + R(p)F(x,px) =0 (p = ^) (1) * "K voprosu o primenenii metoda harmonicheskoi linearizatsii v teorii regulirovaniya." Doklady Akademii Nauk (SSSR), vol. 106, no. 2, 1956, pp. 211-214. 2 WACA TM 1406 where Q(p) and R(p) are polynomials of any degree, of which some re- quired properties will be established in the following paragraphs, and F(x,px) is a given nonlinear function possible only with respect to as- sumptions of the most general character. However, in problems of the theory of control, no assumptions must be made as to the smallness of the nonlinear function F(x,px) or to its small difference from a linear function. In order to render explicit the form in which it would be possible to write a small parameter in equation (l), we shall proceed as follows . Let equation (l) have a periodic solution or a solution approximately periodic differing slightly from the sinusoidal. We write this solution in the form X = X* + ey(t) X = a sin oot (2) where e denotes a small parameter, and y(t) denotes an unknown bounded function of time. In the case of the existence of a periodic solution, we write 00 ey/t) = e^ a^ sin(koDt + ^) (3) k=2 We represent the given nonlinear function F(x,px) in the form F(x,p) = F(x*,px*) + [F(x* + £y, px* + epy) -F(x*,px*)] (4) Expanding the two components separately in a Fourier series, we obtain F(x,px) = q^ + (A + ^ p) sin oot + 2_. F, + &Y2 \ / k=2 k=0 \ (5) k where q^. A, and B are the coefficients of the initial term, the sine, and cosine terms, rexpectively , the cosine being replaced by — p sin oot of the expansion of the function F(x*,px*) in a Fourier series. ZF-^ denotes all the higher harmonics of the expansion of F(x*,px*) in a Fourier series (they must not be considered small, since the nonlinearity is not small) where we write F^ = bj^ sin (koit +^^) (k = 2, 3, . . ., -) (6) e2§, denotes all the terms of the expansion in a Fourier series of all the expressions shown in brackets in formula (4) . This entire expression NACA TM 1406 is written with a small parameter which, according to equation (4), is small if the derivatives Sf/Sx and SF/Spx are finite. This expression is also computed as small in the case of certain discontinuous nonlinear characteristics (e.g., the Raleigh type where the preceding derivatives at the points of discontinuity are delta functions) . We may write e* ^ = ecj^ sin(tojat + ^^) (k = 0, 1, 2, ,-) (7) We substitute equations (2) and equation (4) in the given equation (l) , so that Q(p)x* + Q(p)ey + I \ OP OP R(p)qO + R(P)U + ^ p) sin oDt + R(p)e^ F^ + R(p)£^j^ = (8) k=2 k=0 Since the equation must be satisfied identically, we separately equate to zero all coefficients with the same order harmonic. We note that formula (8) , in the case of the existence of a periodic solution, is exact . From the equating of the zeroth harmonics of equation (8), there is obtained with an accuracy up to e the relation (9) which is a certain general requirement for F(x,px) . From the equating of the first harmoics of equation (8) , taking account of equations (2) and equation (7), we have a sin oot = - A^ A'^ + B^ R(iao) Q(icjo) sin (out + y + p) - ECn R(icu) Q(ia)) sm (oot + ^3_ + P) (10) where x ^^^ P s-^e arguments of the expressions A + iB and R(icD)/Q(iaL)) . On the basis of the exact equation (lO), we obtain the following approximation (with an accuracy up to e): -/- A^ + B^ R(iGD) Q(ico) r + P (11) NACA TM 1406 It is here assumed that the polynomial Q(p) in equation (l) does not have purely imaginary roots . From the equating of the higher harmonics of equation (8), consider- ing equations (3), (s), and (7), we have ■.a^ sin(ka3t + 'Pv) R(ika), Q(ikjD sin(ka)t + \ + P^^)-ecj^ Rfikoj) Q(ika)) sin(koDt + *i^ + Pi^) (12) where ^-^ denotes the argument of the expression R(ikJo)/Q(il2Jo) . It is thus seen that if bv_ is not small, the magnitude I R(ika))/Q(ikco)| should be of the order of e. The last component in equation (12) will then be of the order e.^ . From the exact equation (12) , we obtain the following approximate equation (with an accuracy up to e) : e a^ = b^ R(ika)) Q(ik£ja) \ + Pk ' '^ (13) Comparing formulas (l3) and (ll) we see that, for example, the wish to have in the solution (see eqs. (2) and (3)) (eaj'« (14) leads to the need of satisfying, in the given equation (l), the following requirement: oo L k=2 R(ikcD) Q(ikGD) ■< (a2 + b2) R(ia)) Q(icD) (15) where its satisfying in the concrete system can be checked after cd is obtained. The degree of Q(p) should, in any case, be higher than that of R(p) . A particular case of the general expression (l5) is intuitively the earlier introduced "filter property" of the linear part. Thus, condition (l5) has been obtained and must be satisfied by the coefficients of the given differential equation (l) in order that a pe- riodic solution, if it exists, may be approximately determined in the form of sinusoids in the presence of a "strong" nonlinearity of F(x,px) . NACA TM 1406 The equation for its approximate determination according to equation (s) with the substitution of sin oot = x'Va, assumes the form [q(p) + R(p) (q + 1^ p)] X* = (16) where '2jt ^ ~ a ~ T^ I -^(^ ^^^ ^> ^^ ^°^ u)sin u du F(a sin u, aoo cos u)cos u du (l7) The replacement of equation (l) by equation (16) with its subsequent investigation by the linear methods is called harmonic linearization. This is formally equivalent to following the mode of writing the initial equation (l) with a small parameter, so that Q(p)x + R(p) fq + _i d X + ef (x,p) = (18) where, according to equation (8), we may write ef(x,p) = R(p) 22 ^k + ^ S\ - (> u-S g !- O S rtS > 5 c o OJ CO o .2 2 > w '^ TJ CO ■*** O Eh < rt Z ZiZO S .-, bD g g '^ < CD t; H o. > Z 0:3 "am o 2 2 Sz < rt z :a ■ C ^ ; o , CO O > rt * ! S "1 ' TO .rt ■a ,0. rt ^2 if> ^ rt 01 ij — -M 3 CM 3 B to S oS CO tj o> i-s C; -< V. G 85 S 2 g co" ciJ 73 •2-° m 13 c n bt , S .5 rt Q) t< 4) JS <1) C '-' Sec bc ° 2 n c £ g .S tfl o- .2 -S ■■* en -g rt „ •" O 0) c 0) "> I. rt 3 « en f- 2 •" a 2 o o o- e ? !3 ^ 5 OJ cj ^ o > a 13 ^ o 3 < e o U rt on 3" i Oi CO f- .a j= 13 CJ e 0) <5 O '^ be — §•< a< z ■icO-S SON Z^ ^ > ^ <« CM s-g- T3 ^H CO -^ 2 o> JqS »Ba Z 2 1^ K §..: woo. ^?, u n. tD 0. c« 0. m .2 o Q, (U c a o > •- rt 2 « 3. at! (u S 1^ ""3 ^ -3 £ § ■ o "- ^ ' S.°o ^ il -^ . - °2 g J: 0) c Q) 5 tc o js ■a s*z! ■S "' 13 S < < z to T3 s5 ( r- c: ^ c a g Q) ,j < g (1) S « s < T, n U r 1 X >. H u U. to > W •o I/l < W ID . o-sw K o o ^ >> U-S g K 2 o S ._ bO N c^ a o a. a CO , u ^ 4) -^j -Q "^ ;::3 CO e^ (ti .2-0 to " 2 -5 ♦^ O 01 c dj ■" t, cij s < CD Z zzo 3 C ;3 W.S S H a > z o- "am Q §13 O 2 2 K a C CO -— • o OS < cd uz z :a . a> cij (U i! ■o m o • o -»- to — • Cl! E c« " 2 ^ -1 -^ 13 W ? a! be . C .5 CJ 4) tU C *J Sec m 3 c S " m -. 3 B & a e ? b 1) *- 5 ") c« ^ 5a°°S, .S a 13 « < ':5 S u a ' a i" ' si o H 3 o ^ _ r! -S m IS'" 2 Sfi a lu CJ ^ ^ CJ g&5 > '-' rt M J- cr 0,3 *(" q; ji: n U rn n n <: b: S 0. z "2 en ■= ■gco'^ z-^ ^ 1-g^ci. ^ -H CO ^q2 o -CM (U ■ 3 o •§ (SS o < (^ Z So CO ^ ■g CO i-i d eg S^a •S S - XJ ^H CO :Sd2 ^oB W CSl ,aB z H ■ ,^ i^ <^ § §0 2; ■> w -^ T3 M g<; p ^ B PS < c< Z zzo 3h m • H d) Q, T -e O c W « o &: s„ o ^ > G 2 o g E 3 0:gS? 5 = :3 < d) Jj W.§ S 33 C - H a > °:a 3 m 3 § CO -^ o « ■» m H^ < cH UZ i< d Z:a dJ eg' ?! m , u .2 2 c o ^ o i I d) -M J3 — .s s :^ .2 S == ? ^ M o a •^ ..^ J3 d !3 d 15 Sam m d ' -i o d ■" >.a ii ■° m a c d m 11^ o c <= 2 2 -3 O d) 3 "> t. d d) d -f- J3 d> ^ 5 S d) •" a a o _ C t, c d) d g- 5^ Q §13 o 2 2 a a t. H 9 d w^ S o o S "> <" -2 2 I ^ ^ a d) d) d " c 5 S .2 ■& CO a, J- (u ^ j2 t-i _ ire "O .-K J^ J£ Cm -^ O • CJ 2 _. oQ -^ CO :j :;i o > (1) >» a m In bb.5' O ^ a O -3 O = ■§3 cn;3 ^ o-9x; en J3 .^ -H 3 CM 3 a .c o-S o g m _ i! > £2 §2 H 3 o a as- si c a o dj -^ ■ o ••3bc5 Sag •H J3 O , « -^ c • O «-■»-, ■ §■13 V-i *^ S ■ o o d ■ d; ■" a d; c d) h O J3 ^§a. a"|- o S ^■ m dj -■ _ c m ^ ■^^ < < ZtH O « SO to M SB 2SS « J z o Z ,, ' 2 ■> 5 KO rR w 3 5S ^ J5 f^ So g cm' a-H • ■-1 o ti o oa,i2 c o a -CD --y u -Jo aj d sir}. M d) ,t! <; d „-3 U Z .-. bO>a i^ 2 g j= 2 -o » a o C y d g d m H o * Q. eg 2 a ' 5r"S — a " d) d) « jq 13 '^ a bO C J3 gco , 2; V W 2 '-' T3 M a . CM o^St- 3 (0 OS ."^ ' is ^ 3 a 'o ba S e« t. e3> Sr3 n i; ^ d) 2 2 -S en = J a d) :a d) d en c 3 „ _ °-^s 2 asl t, 'S, m a d) 3 §.&S-c_§^o 2.S°£&5.2 m _ i! > '-' d t. H *j CJ at3 dj Q g d) d g > d) s s ■a 5 O s-i t. o a^ ft e« ° a dj d} 3 b o bOrl S a ^ 3 a) tT ^> o S « di j= = ti d) ?5 o ^ .» o •^5 aS QJ O j3 2 = 3 o ;^ ^^ H Si < < z o o q cil I o S » 1 6 to to >> 3 " o Si3 5 « f-t s bO.S < O QJ *< < s? s S Id < < 2 S < 5 s 3 s < < z o o u u s I CO « S "S « «> < o I e o u u •■3 s I "S n s « 3 £ "53 m -^ 5 S < z bo .3 l§ X C (U C4 < < Z bo.a I = ■o o c c >< S rf (u S . c o o 1 Rese Math Popo NAC Akad Dokl 1956 «dB CM 2 03 ► CM K'^ ,. • s ■ < ii rt . 0) -; ay c -s a Mia o 5 0) <5 o •< OS S ei, z ^S2 ^bH o o H <: u 2 o h a o u CI I e "3 ii V CO 3 & i :§ 1 § B c £ ■a x B S ol < s < < a o o u i s I a at 01 c "3 "S CD 3 S' 1 13 0) hn .s r" t1 o c c < < 2 H o < 1 3 "S B S CO 3 s-