Derivation of Robust Stability Ranges for Disconnected Region with Multiple Parameters

SICE Journal of Control, Measurement, and System Integration, Vol. 10, No. 1, pp. 03 038, January 017 Derivation of Robust Stability Ranges for Disconnected Region with Multiple Parameters Tadasuke MATSUDA Abstract : The aim of this paper is to give an extension of the paper [T. Matsuda et al. Proc. 33rd IASTED Modelling, Identification and Control, 809-004, 014], which gives a robust stability condition for a system with disconnected stability regions. The considered system depends on only one uncertain parameter. In this paper, an explicit algorithm to derive the stability ranges for disconnected stability regions is given. We also extend the result to the case that the system depends on multiple uncertain parameters. A numerical example shows that the proposed method can be applied to robust stability analysis of the lateral dynamics of an aircraft even if all the coefficients of the characteristic polynomial vary. The numerical example also shows that the stability ranges derived by the proposed method are larger than those by a former method. Key Words : robust control, stability analysis, parameter space, aircraft, stability feeler. 1. Introduction Robust stability is an important characteristic for control systems with uncertainties because of the difference between the models and actual systems. Robust stability analysis methods have been developed especially based on a parametric approach [1] [3]. Kharitonov s theorem [4] gives a condition for robust Hurwitz stability of interval polynomials. Edge theorem shown in [5] implies that a polytope of polynomials is stable if and only if all the segment polynomials corresponding to the exposed edges are stable. A method for Hurwitz stability analysis of a segment polynomial is given in [6]. Schur stability of a segment polynomial is discussed in [7]. A method for stability analysis of delta-operator segment polynomial is presented in [8]. On the other hand, it is also required to obtain stability ranges of uncertain parameters in the case of design of a robust system, where stability ranges mean a subset of the ranges of uncertain parameters stabilizing the systems. There are some methods and tools to meet such a demand. Methods in [9] [14] are based on Lyapunov stability theory, which can derive the stability ranges. A method in [15] is based on the result given in [6]. Methods based on the guardian maps are given in [16] [18]. A method using LMI formulation is shown in [19]. The stability radius [0],[1] and the directional stability radius [] are tools to derive the stability ranges of uncertain systems. The stability feeler [3] also meets such a demand, which enables one to derive the stability ranges of a segment polynomial. The stability feeler can be applied to many types of stability analysis if the stability region in the complex plane is connected. Recently the presented conference paper [4] gives a robust stability condition based on the stability feeler for the case of Department of Electrical and Electronic Engineering, Faculty of Engineering, Chiba Institute of Technology, 17 1 Tsudanuma, Narashino, Chiba 75 0016, Japan E-mail: tadasuke.matsuda@p.chibakoudai.jp (Received February 3, 016) (Revised October 8, 016) disconnected stability regions in the complex plane. Stability analysis for disconnected stability regions is required as shown in [19],[5]. The advantage of using the stability feeler is that we can derive the exact stability ranges if the system depends on one uncertain parameter. If the system depends on multiple parameters, we can derive almost exact stability ranges by using the stability feeler, although they are not the exact stability ranges. Almost exact stability ranges are ranges which are not exact but can be considered to be exact practically 1. Here, by stability we mean having all zeros of characteristic polynomials in prescribed regions in the complex plane. The phrase stability region means the above prescribed region in the complex plane, which may be disconnected. Therefore, stability in this paper includes performance of the system. This is a reason why stability analysis for the case of disconnected stability regions is required. An application to the lateral dynamics of an aircraft is also shown in the conference paper. However, an explicit algorithm to derive the stability ranges is not given in [4]. Moreover, it is assumed that uncertainty exists only in one term of the characteristic polynomial of the lateral dynamics of an aircraft. The aim of this paper is to give extensions of [4]. An explicit algorithm to derive the stability ranges for disconnected stability regions is given. The case that characteristic polynomials depend on multiple uncertain parameters is also considered. The method presented in [3] enables one to derive the stability ranges. However, the method in [3] cannot be applied if the stability region in the complex plane is disconnected. In [4], systems with only one parameter are considered. In [6], a stability condition of systems for disconnected stability regions in the complex plane is given. However, methods to derive the stability range are not given in [6]. Contrary to them, the paper gives an algorithm to derive the stability ranges of systems with multiple parameters when the stability regions in the complex plane are disconnected. Table 1 shows what this paper deals with. A numerical example shows that the proposed method 1 Details are shown in Remark of this paper. JCMSI 0001/17/1001 003 c 016 SICE

Table 1 What this paper deals with. stability types one parameter multiple parameters connected region [3] [3] disconnected regions [4], This paper This paper SICE JCMSI, Vol. 10, No. 1, January 017 33 Fig. r i and c i in the complex plane. Fig. 1 Disconnected stability regions. can be applied to stability analysis of the lateral dynamics of an aircraft even if all the coefficients of the characteristic polynomial vary. This paper also compares the result by the proposed method and that by the method in [19]. Stability ranges derived by the proposed method turn out to be larger than those given in [19]. The notations used in the paper are as follows. R: the set of real numbers. R p :thesetofp-dimensional real column vectors. R p q :thesetofp-by-q real matrices. j: the imaginary unit, i.e., j = 1. A T : transposition of matrix A.. Robust Stability Condition In this section, a necessary and sufficient condition for robust stability given in [4] is shown. We consider a real characteristic polynomial family given by f (s) + ag(s) a R, (1) where f (s) = n f i s i, f n 0andg(s) = n g i s i are fixed real polynomials. It is assumed that both of the leading coefficients of f (s) andg(s) are greater than zero without loss of generality. It is also assumed that the degree of (1) does not drop. A characteristic polynomial family is said to be robustly stable if all the roots of all the polynomials in the polynomial family are in the prescribed open stability region in the complex plane. It is considered that the stability region in the complex plane can be disconnected. It is assumed that the number of connected regions of which the stability regions in the complex plane consist and that of real numbers on the boundary of the stability regions in the complex plane are finite. We can also assume that the stability regions are symmetric with respect to the real axis without loss of generality because considered characteristic polynomials are real polynomials. An example of the stability regions satisfying the above assumptions is shown in Fig. 1. To state the main result in [4], we define a i, i = 0, 1,..., m as follows. Define a 0 := f n /g n if g n 0. If g n = 0, define a 0 :=. Leta 1 a a m be all the stability-boundarycrossing points greater than a 0, which implies that there exists a root of polynomial f (s) + a i g(s) on the stability boundary in the complex plane for all i = 1,,...,m. Definea m :=+. Then the following robust stability condition is given: Lemma 1. [4] All the polynomials in set f (s) + ag(s) a (a i, a i ) for a given integer i 0, 1,,..., m are stable if and only if there exists a stable polynomial in f (s) + ag(s) a (a i, a i ) for the same i. The following lemma gives a method to derive the above stability-boundary-crossing points a 1,...,a m. Lemma. [4] The set of stability-boundary-crossing points a 1,...,a m equals the union of sets a a > a 0, e T x ( f + ag) = 0, x i r R, i = 1,,...,β () and a a > a 0, E x+ jy ( f + ag) = 0, x + jy i c, x, y R, y > 0, i = 1,,...,β. (3) The notations used in Lemma are as follows: Coefficient vectors f R n and g R n are defined as f := [ f 0 f 1 f n ] T, (4) g := [g 0 g 1 g n ] T, (5) respectively. A vector e x R n and a -by-(n + 1) real matrix E x+ jy R (n) are defined as e x := [1 xx x n ] T, (6) E x+ jy := h1 x+ jy h, (7) x+ jy respectively, where h 1 x+ jy := [h 0 h 1 h n ], (8) h x+ jy := [0 h 0 h 1 h n 1 ], (9) h i := xh i 1 (x + y )h i, i =,...,n, (10) h 0 := 1, h 1 := x. (11) i r and i c, i = 1,,...,β are sets of real numbers and complex numbers on the boundary of connected region S i, i = 1,,...,β, respectively, where S i are connected regions of which the given disconnected stability regions in the complex plane consist. Figure shows an example of i r and i c in the complex plane. 3. Results This section gives an algorithm to derive stability ranges based on Lemmas 1 and. We also extend the result to the multiple-parameter case. Finally, a numerical example is shown to compare the result by the proposed method and that by the method in [19]. 3.1 Algorithm to Derive Stability Ranges We give an algorithm to obtain the ranges of a such that the polynomial family (1) is robustly stable based on [4]. The stability regions in the complex plane can be disconnected. The

34 SICE JCMSI, Vol. 10, No. 1, January 017 Algorithm 1 Derivation of stability range one parameter case 1: Derivation of boundary-crossing points: fn /g n, g n 0 : a 0 :=, g n = 0 3: A r i := a a > a 0, e T x ( f + ag) = 0, x i r R, i = 1,,...,β. 4: A c i := a a > a 0, E x+ jy ( f + ag) = 0, x + jy i c, x, y R, y > 0, j = 1, i = 1,,...,β. β 5: Define a 1 a m as all the elements of set (A r i Ac i ). 6: Define a m :=+. 7: Derivation of stability ranges by using the boundary-crossing points: 8: for i = 0 to m do 9: if there exists a (a i, a i ) s.t. f (s) + ag(s)isstablethen 10: f (s) + ag(s) is robustly stable for a (a i, a i ) 11: else 1: f (s) + ag(s) is not stable for all a (a i, a i ) 13: end if 14: end for algorithm is given as Algorithm 1. 3. Multiple-Parameter Case In this subsection we extend the above result to the multipleparameter case. We consider the problem of determining ranges of a real uncertain parameter K robustly stabilizing the following characteristic polynomial family with multiple uncertain parameters: P = ˆp(s) + α i p i (s) + Kp K (s) α i [ r i, r i ] R, i = 0,...,l, K R, (1) where ˆp(s), p i (s) andp K (s) are fixed real polynomials. We assume that the degree of P does not drop and the highest coefficient of P is positive for all α i [ r i, r i ], i = 0,...,l when K = 0. In the case of connected stability region in the complex plane, a method to solve the above problem is given in [3] as shown in Table 1. If Edge theorem [5] is satisfied for the case of disconnected stability regions, the method in [3] can also be applied to such a case. Fortunately, Edge theorem is satisfied for a class of disconnected stability regions as follows: Lemma 3. [6] Assume that the complement of stability regions in the complex plane S i β is pathwise connected on the Riemann sphere. Then, polynomial family P is robustly stable if and only if its exposed edges are stable. From the above lemma, one can derive stability ranges of K by the same method in [3] if the complement of stability region in the complex plane is pathwise connected. The following theorem is therefore satisfied: Theorem 1. Assume that the complement of stability regions in β the complex plane S i is pathwise connected on the Riemann sphere. Then, polynomial family P is robustly stable for all K satisfying K ([K i 1 1 +Δ, Ki 1 Δ] [Ki 1 +Δ, Ki Δ] [K i b 1 +Δ, Ki b Δ]), i 1, i,..., i b I, (13) Algorithm Derivation of stability ranges multiple-parameter case 1: Let (K 1 1, K1 ) (K 1, K ) (Kb 1, Kb ) be the set of all the values of K ( M, + M) which stabilize ˆp(s) + r i± p i (s) + Kp K (s) r i± r i, r i, i = 1,..., l, where M < + is a sufficiently large real number and K 1 1 < K1 K 1 < K Kb 1 < Kb. : I := 1,,..., b. 3: repeat 4: if there exist α e [ r e, r e ], r i± r i, r i, c 1,, d I, e 0,..., l s.t. ˆp(s) + r i± p i (s) + r i± p i (s) + (Kc d Δc)p K (s) + α e p e (s) i=e is unstable, where Δ > 0issufficiently small real number then 5: if there exists K (K d 1, Kd ) s.t. ˆp(s) + r i± p i (s) + r i± p i (s) + α e p e (s) + Kp K (s) i=e is stable then 6: Let ( K d 1, K d ) (Kb 1, Kb ) (Kb+b 1, K b+b )bethe set of these values, where K d 1 < K d Kb 1 < K b K b+b 1 < K b+b. 7: K d 1 := K d 1, Kd := K d, I := I b + 1,..., b + b. 8: else 9: I := I\d 10: end if 11: end if 1: until ˆp(s) + r i± p i (s) + r i± p i (s) + (Kc d Δc)p K (s) + α e p e (s) i=e is stable for all α e [ r e, r e ], r i± r i, r i, c 1,, d I, e 0,..., l. Then stability ranges of K are given as [K i1 1 +Δ, Ki1 Δ] [Ki 1 +Δ, Ki Δ] [Ki b 1 +Δ, Ki b Δ], where i 1, i,..., i b I. and there exists an unstable polynomial in P for all K satisfying K ([ M +Δ, K i 1 1 ] [Ki 1, Ki 1 ] [Ki, Ki 3 1 ] [K i b, M Δ]), i 1, i,..., i b I, (14) where K i 1 1 < Ki 1 Ki 1 < Ki < Ki b 1 < Ki b and I are derived by Algorithm. Remark 1. The stability ranges of polynomials given in the lines 1, 4 and 5 of Algorithm are determined by Algorithm 1. Remark. The stability ranges (13) are almost exact. The word almost is used above because we cannot know whether P is robustly stable for K (, M +Δ), (K d 1, Kd 1 + Δ), (K d Δ, Kd ), (M Δ, ), or not by the algorithm, where d I. Figure 3 shows an example of a derivable stability range by the proposed method when I = d. However, the obtained stability range can be considered to be exact practically because M and Δ > 0 can be set to be sufficiently large and small, respectively. Since Δ can be set to be sufficiently small, the ranges (K d 1, Kd 1 +Δ), (Kd Δ, K d ), d I, which cannot be known whether stability ranges or not, can be sufficiently small as shown in Fig. 3. By the proposed method we can know the stability ranges in [ M + Δ, M Δ] instead of (, ), except for the small ranges (K d 1, Kd 1 +Δ), (Kd Δ, Kd ), d I. The range [ M + Δ, M Δ] can be enlarged as shown in Fig. 3, because we can set M and Δ to be sufficiently large and small, respectively.

SICE JCMSI, Vol. 10, No. 1, January 017 35 Fig. 3 An example of a derivable stability range by the proposed method. Since this method is for robust stability test of polynomials with scalar uncertainty, a numerical check by gridding the parameter space can be an alternative way. However, the method using gridding has a risk to miss unstable polynomials. On the other hand, the proposed method does not have such a risk because the stability condition is sufficient (necessary and sufficient when M and Δ are sufficiently large and small, respectively). Note that our method cannot be applied if the complement of the stability region in the complex plane is not pathwise connected. The example in the next subsection satisfies the condition that the complement is pathwise connected. 3.3 Robust Stability Analysis of an Aircraft In this subsection, robust stability analysis for the lateral dynamics of an aircraft taken from [19],[5] is given. This example shows that one can derive the stability range of an uncertain parameter by using Algorithm. The example also shows that the stability ranges derived by the proposed method are larger than those given in [19]. The nominal characteristic polynomial of the lateral dynamics of an aircraft is given as ˆp(s) := 5 + 154s + 1.5s + 50s 3 + 10.5s 4 + s 5. (15) The zeros of ˆp(s)are 0.5, ± j, 3 ± j. (16) Unlike in [4], we consider the case that all the coefficient of the characteristic polynomial varies. The characteristic polynomial family is given by 4 P = ˆp(s) + α i p i (s) + Kp K (s) α i [ r i, r i ] R, (17) where p i (s) := s i, i = 0, 1,, 3, 4, p K (s) := s 5, r 0 := 5 0.01k, r 1 := 154 0.001k, r := 1.5 0.00k, r 3 := 50 0.007k, r 4 := 10.5 0.0k. (18) Fig. 4 The stability region of the example in the complex plane. This characteristic polynomial family has six uncertain parameters: α 0, α 1, α, α 3, α 4 and K. The stability region in the complex plane is inside the disks of radius 0.4 around the zeros of ˆp(s) as shown in Fig. 4. The stability region satisfies the condition that its complement is pathwise connected on the Riemann sphere. If we assume that K [ 1 0.05k, 1 0.05k], the polynomial family P given in (17) coincides with the polynomial family considered in Example 1 of [19]. In [19], it is shown that the system is robustly stable if k 0.41. In this paper, we show that the proposed method enables one to derive a larger stability range. Now we define that k := 0.5 and derive ranges of K stabilizing P. First, we execute the line 1 of Algorithm. In order to derive K stabilizing 4 ˆp(s) + r i± p i (s) + Kp K (s) r i± r i, r i, (19) we derive K stabilizing the following 5 polynomials: ˆp(s) r 0 p 0 (s) r 1 p 1 (s) r p (s) r 3 p 3 (s) r 4 p 4 (s) + Kp K (s) ˆp(s) + r 0 p 0 (s) r 1 p 1 (s) r p (s) r 3 p 3 (s) r 4 p 4 (s) + Kp K (s) ˆp(s) r 0 p 0 (s) + r 1 p 1 (s) r p (s) r 3 p 3 (s) r 4 p 4 (s) + Kp K (s). ˆp(s) + r 0 p 0 (s) + r 1 p 1 (s) + r p (s) + r 3 p 3 (s) + r 4 p 4 (s) + Kp K (s) (0) by Algorithm 1 and obtain the intersection of each stability range. Stability-boundary-crossing points of the first polynomial of (0) are derived as follows. Define f (s) := ˆp(s) r 0 p 0 (s) r 1 p 1 (s) r p (s) r 3 p 3 (s) r 4 p 4 (s), g(s) := p K (s). (1) The coefficient vectors are then given by f = [51.87 153.9615 1.43875 49.915 10.4475 1] T, () g = [0 0 0 0 0 1] T. (3)

36 SICE JCMSI, Vol. 10, No. 1, January 017 Therefore a 0 is defined as a 0 = f 5 /g 5 = 1/1 = 1 inthe line of Algorithm 1. The connected regions of which stability region in the complex plane consist are given as S 1 := x + jy (x + 0.5) + y < 0.4, (4) S := x + jy (x + ) + (y ) < 0.4, (5) S 3 := x + jy (x + ) + (y + ) < 0.4, (6) S 4 := x + jy (x + 3) + (y ) < 0.4, (7) S 5 := x + jy (x + 3) + (y + ) < 0.4, (8) asshowninfig.4,wherex and y are real numbers. Then, the set of points on the boundaries are given as 1 r = 0.1, 0.9, (9) i r = φ, i =, 3, 4, 5, (30) 1 c := x + jy (x + 0.5) + y = 0.4, y 0, (31) c := x + jy (x + ) + (y ) = 0.4, (3) 3 c := x + jy (x + ) + (y + ) = 0.4, (33) 4 c := x + jy (x + 3) + (y ) = 0.4, (34) c 5 := x + jy (x + 3) + (y + ) = 0.4, (35) where φ is the empty set. Since 1 r = 0.1, 0.9, we derive the values of a which satisfies e T x ( f + ag) = 0forthe case that x = 0.1 andx = 0.9 in order to derive A r 1 in the line 3 in Algorithm 1. In the case that x = 0.1, the equation e T x ( f + ag) = 0 has the solution a = 3.765 10 6. (36) In the case that x = 0.9, the equation e T x ( f + ag) = 0hasthe solution a = 9.88. (37) However, 9.88 A r 1 because 9.88 < a 0 = 1. A r 1 is therefore determined as A r 1 = 3.765 106. (38) On the other hand, A r i = φ, i =, 3, 4, 5 (39) because i r = φ, i =, 3, 4, 5. The sets A c i givenintheline4 in Algorithm 1 is derived as follows. To derive A c 1,wesolvethe following system of equations E x+ jy ( f + ag) = 0 (40) (x + 0.5) + y = 0.4. The solution of the above is obtained as a = 1.14 10 6. (41) However, 1.14 10 6 A c 1 because 1.14 106 < a 0.We therefore obtain A c A c 1 = φ. (4) can be derived by solving E x+ jy ( f + ag) = 0 (x + ) + (y ) = 0.4. (43) The solutions of the above are obtained as a = 0.056, 0.080. (44) Both of the solutions of a is greater than a 0 = 1. Therefore A c is given as A c = 0.056, 0.080. (45) Similarly, A c 4 is derived as A c 4 = 0.0445, 0.03. (46) It is obvious that A c 3 = Ac 5 = φ (47) because there are no y > 0 satisfying x + jy i c for 5 i = 3 or 5. Form the above, set (A r i A c i ) is equal to 0.056, 0.0445, 0.03, 0.080, 3.765 10 6. Therefore the stability-boundary-crossing points are given as a 1 = 0.056, (48) a = 0.0445, (49) a 3 = 0.03, (50) a 4 = 0.080, (51) a 5 = 3.765 10 6. (5) Now we derive the stability ranges of the first polynomial of (0) by using the above stability-boundary-crossing points. One can see that f (s) + a 0+a 1 g(s), f (s) + a 1+a g(s), f (s) + a 3 +a 4 g(s), f (s) + a 4+a 5 g(s)and f (s) + (a 5 + 1)g(s) are not stable and that f (s) + a +a 3 g(s) is stable. It is concluded that the first polynomial of (0) is stable for all K (a, a 3 ) = ( 0.0445, 0.03). (53) The stability ranges of all the polynomials of (0) can be derived by the same way. The intersection of these stability ranges is given by (K 1 1, K1 ) = ( 0.0170, 0.0181), (54) which is equal to the range of K stabilizing (19). Now we execute the lines 1 of Algorithm. Stability of ˆp(s) + r i± p i (s) + r i± p i (s) + (Kc d Δc)p K (s) i=e + α e p e (s) (55) needs to be examined. It is possible by using Algorithm 1. As a result, the polynomial is stable for all α e [ r e, r e ], r i± r i, r i, c 1, + 1, d I, e 1,..., l, wherei = 1 and Δ=1 10 4. Therefore the range of K stabilizing the system is given by [ 0.0170 +Δ, 0.0181 Δ] = [ 0.0169, 0.0180]. (56) Now let us compare the result in [19] and that by the proposed method. The result in [19] is as follows. The system is stable if k 0.41, i.e., ˆp(s) + 4 α i p i (s) + Kp K (s) (57)

SICE JCMSI, Vol. 10, No. 1, January 017 37 is stable for all α 0 [ 0.153, 0.153], α 1 [ 0.037114, 0.037114], α [ 0.059045, 0.059045], α 3 [ 0.08435, 0.08435], α 4 [ 0.05061, 0.05061], K [ 0.0105, 0.0105]. (58) On the other hand, the result by the proposed method shows that (57) is stable for all α 0 [ 0.13, 0.13], α 1 [ 0.0385, 0.0385], α [ 0.0615, 0.0615], α 3 [ 0.0875, 0.0875], α 4 [ 0.055, 0.055], K [ 0.0169, 0.0180]. (59) The obtained stability ranges are larger than those shown in [19]. 4. Conclusion This paper has given extensions of [4]. The algorithm shown in Subsection 3.1 enables one to derive stability ranges for disconnected stability region. We also have shown another algorithm to derive stability ranges in the case that characteristic polynomials depend on multiple uncertain parameters in Subsection 3.. The numerical example shows that the proposed method can be applied to stability analysis of the lateral dynamics of an aircraft even if all the coefficients of the characteristic polynomial vary. It is shown that stability ranges derived by the proposed method are larger than those given in [19]. Acknowledgments The author thanks Mr. Takanao Matsui and Prof. Toshiaki Yachi for discussion. He also thanks reviewers and the Associate Editor for helpful advice. References [1] B. Barmish: New Tools for Robustness of Linear Systems, Macmillan Coll Div, 1994. [] S. Bhattacharyya, H. Chapellat, and L. Keel: Robust control: The parametric approach, Prentice Hall, 1995. [3] J. Ackermann: Robust Control: Systems with Uncertain Physical Parameters, Springer-Verlag, 1993. [4] V.L. Kharitonov: Asymptotic stability of an equilibrium position of a family of systems of differential equations, Differentsialnye Uravneniya, Vol. 14, No. 11, pp. 1483 1485, 1979. [5] A.C. Bartlett, C.V. Hollot, and H. Lin: Root locations of an entire polytope of polynomials: It suffices to check the edges, Math. Contr., Signals, and Syst., Vol. 1, No. 1, pp. 61 71, 1988. [6] S. Bialas: A necessary and sufficient condition for the stability of convex combinations of stable polynomials or matrices, Bulletin of the Polish Academy of Sciences: Technical Sciences, Vol. 33, No. 9 10, pp. 473 480, 1985. [7] J.E. Ackermann and B.R. Barmish: Robust Schur stability of a polytope of polynomials, IEEE Trans. Autom. Contr., Vol. 33, No. 10, pp. 984 985, 1988. [8] T. Matsuda, K. Kawabata, and T. Mori: Stability analysis methods for delta operator polytopic polynomials, Electrical Engineering in Japan, Vol. 159, No. 3, pp. 56 64, 007. [9] R.V. Patel and M. Toda: Quantitative measures of robustness for multi-variable systems, Proc. Joint Automatic Control Conference, San Francisco, TP8-A, 1980. [10] R.K. Yedavalli: Improved measures of stability robustness for linear state space models, IEEE Trans. Autom. Contr., Vol. AC- 30, No. 6, pp. 557 579, 1985. [11] K. Zhou and P.P. Khargonekar: Stability robustness bounds for linear state-space models with structured uncertainty, IEEE Trans. Autom. Contr., Vol. AC-3, No. 7, pp. 61 63, 1987. [1] D.S. Bernstein and W.M. Haddad: Robust stability and performance analysis for linear dynamic systems, IEEE Trans. Autom. Contr., Vol. AC-34, No. 7, pp. 751 758, 1989. [13] D.S. Bernstein and W.M. Haddad: Robust stability and performance analysis for state space systems via quadratic Lyapunov Bounds, SIAM J. Matrix Anal. Appl., Vol. 11, No., pp. 39 71, 1990. [14] R.K. Yedavalli: Robust root clustering for linear uncertain systems using generalized Lyapunov theory, Automatica, Vol. 9, No. 1, pp. 37 40, 1993. [15] M. Fu and B.R. Barmish: Maximal unidirectional perturbation bounds for stability of polynomials and matrices, Systems & Control Letters, Vol. 11, No. 3, pp. 173 179, 1988. [16] S. Rern, P.T. Kabamba, and D.S. Bernstein: Guardian map approach to robust stability of linear systems with constant real parameter uncertainty, IEEE Trans. Autom. Contr., Vol. 39, No. 1, pp. 16 164, 1994. [17] L. Saydy, A.L. Tits, and E.H. Abed: Maximal range for generalized stability: Application to physically motivated examples, Int. J. Contr., Vol. 53, No. 4, pp. 837 845, 1991. [18] X. Zhang, P. Tsiotras, and A. Lanzon: An approach for computing the exact stability domain for a class of LTI parameter dependent systems, Int. J. Contr., Vol. 79, No. 9, pp. 1046 1061, 006. [19] Y. Dolgin and E. Zeheb: LMI characterization of general stability regions for polynomials, IEEE Trans. Autom. Contr., Vol. 56, No. 4, pp. 890 895, 011. [0] C.B. Soh, C.S. Berger, and K.P. Dabke: On the stability properties of polynomials with perturbed coefficients, IEEE Trans. Autom. Contr., Vol. 30, No. 10, pp. 1033 1036, 1985. [1] L. Qiu, B. Bernhardsson, A. Rantzer, E. Davison, P. Young, and J. Doyle: A formula for computation of the real stability radius, Automatica, Vol. 31, No. 6, pp. 879 890, 1995. [] K. Kawabata, T. Mori, and Y. Kuroe: Directional stability radius: A stability analysis tool for uncertain polynomial systems, IEEE Trans. Autom. Contr., Vol. 48, No. 6, pp. 101 1016, 003. [3] T. Matsuda and T. Mori: Stability feeler: A tool for parametric robust stability analysis and its applications, IET Contr. Theory Appl., Vol. 3, No. 1, pp. 165 1633, 009. [4] T. Matsuda, T. Matsui, and T. Yachi: Application of the stability feeler to lateral dynamics of an aircraft with disconnected regions, Proc. 33rd IASTED Int. Conf. Model. Identif. Contr., 809-004, 014. [5] O. Bachelier, D. Henrion, B. Pradin, and D. Mehdi: Robust root-clustering of a matrix in intersections or unions of regions, SIAM J. Contr. Optim., Vol. 43, No. 3, pp. 1078 1093, 004. [6] M. Fu and B.R. Barmish: Polytopes of polynomials with zeros in a prescribed region, IEEE Am. Contr. Conf., 1988, pp. 461 464, 1988.

38 SICE JCMSI, Vol. 10, No. 1, January 017 Tadasuke MATSUDA (Member) He received his B.S., M.S., and Ph.D. degrees from Kyoto Institute of Technology, Japan, in 003, 005, and 008, respectively. In 016, he joined Chiba Institute of Technology, where he is currently an Associate Professor of the Department of Electrical and Electronic Engineering, Faculty of Engineering. His research interests include robust stability theory and power systems. He is a member of ISCIE and IEEJ.