Advanced
A Tuning Algorithm for LQ-PID Controllers using the Combined Time - and Frequency-Domain Control Method
A Tuning Algorithm for LQ-PID Controllers using the Combined Time - and Frequency-Domain Control Method
Journal of Electrical Engineering and Technology. 2015. May, 10(3): 1244-1254
Copyright © 2015, The Korean Institute of Electrical Engineers
This is an Open-Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/)which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
  • Received : August 12, 2014
  • Accepted : December 29, 2014
  • Published : May 01, 2015
Download
PDF
e-PUB
PubReader
PPT
Export by style
Article
Author
Metrics
Cited by
TagCloud
About the Authors
Chang-Hyun Kim
The Research Institute of Industrial Science, Hanyang University, Korea. (chang1005@hanyang.ac.kr)
Ju Lee
Dept. of Electrical Engineering, Hanyang University, Korea. (julee@hanyang.ac.kr)
Hyung-Woo Lee
Corresponding Author: Dept. of Railway Vehicle System Engineering, Korea National University of Transportation, Korea. (krhwlee@ut.ac.kr)

Abstract
This paper proposes a new method for tuning a linear quadratic - proportional integral derivative controller for second order systems to simultaneously meet the time and frequency domain design specifications. The suitable loop-shape of the controlled system and the desired step response are considered as specifications in the time and frequency domains, respectively. The weighting factors, Q and R of the LQ controller are determined by the algebraic Riccati equation with respect to the limiting behavior and target function matching. Numerical examples show the effectiveness of the proposed LQ-PID tuning method
Keywords
1. Introduction
Linear optimal control theory has been extensively studied since Bellman’s dynamic programming [1] and Pontryagin et al.’s minimum principle [2] were published early in the 1960s. In linear optimal control, the linear quadratic regulator (LQR) problem is to find the optimal feedback control law to minimize the quadratic performance criterion with respect to the states and inputs via weighting matrices Q and R [3 - 5] . Thus the central issue remains of how to relate the weighting matrices in the quadratic performance criterion to classical specifications in the time and frequency domains for the control system design [6 - 9] .
Various analytic methods have been developed for frequency domain specifications such as disturbance attenuation, noise rejection, and good command following for the LQR [10 , 11] and linear quadratic Gaussian / linear transfer recovery (LQG/LTR) [12 , 13] , based on the loopshaping method. These existing methods have difficulty directly matching the quadratic performance indices with the filter dynamics to the frequency loop shape [3 , 4] .
On the other hand, eigenvalue positioning methods were proposed for the time domain approach to select the appropriate weighting matrices, Q and R, to locate the closed loop poles inside the desired region or match them with the specific eigenvalues through the ARE [4] . Existing pole positioning approaches are limited in their analysis of time domain performances such as overshoot, rising and settling time [4] . LQ-proportional integral (PID) control methods have been also developed separately in time and frequency domains [14 , 15] .
We propose a new optimal control method for LQ-PID design to simultaneously satisfy the design specifications in the time- and frequency-domain without any filter or pole-zero cancelation [16] . For frequency-domain performances, we improved the existing limiting behaviour method of the loop transfer function to more exactly interpret the loop shape with respect to all design factors, based on Kalman’s equality [17 , 18] . This loop shaping makes it possible to consider the magnitudes of the LQ loop system independently in low and high frequency ranges for good command following, disturbance rejection, and attenuation of noise effect [19] . Time domain specifications such as percent overshoot, rise time, and settling time are related to the design factor by target function matching regardless of the pole-zero cancellation [16] . Overshoot and the Routh- Hurwitz stability criterion are formulated by the coefficient analysis with respect to the normalized third-order closed-loop system [20] . The other time-related specifications are addressed by the timeregulation property of the target function [21] . In the proposed method, the LQ-PID control parameters are determined by optimizing to minimize the cost function, which is subject to constraints to meet the combined time - and frequency-domain specifications for classical performances. We choose a combination of integral absolute error (IAE) and 2-norm of the control gains for the cost function of the optimal problem [22] .
The primary contributions of the proposed method include as follows:
  • (i) The proposed method provides the direct matching manner to relate the weighting factor, Q, to the loop shape of the loop transfer function in frequency domain.
  • (ii) The proposed target function matching method improves the existing time domain approach of LQPID control by considering the specific time domain performances of transient response such as overshoot, rising time, settling time.
  • (iii) Both of time and frequency domain specifications are combined by the proposed optimization problem in order to be simultaneously satisfied.
The paper is organized as follows. In Section 2, we introduce the LQ-PID formulation to design the optimal PID controller by transforming the PID control into the LQ approach. The design methodologies for the proposed controller are discussed as the optimization problem with constraints subject to the time- and frequency-domain specifications in Section 3. Numerical examples are given in Section 4 to show the effectiveness of the proposed design method, compared to other design methods. Finally, Section 6 contains some concluding remarks and remaining challenges.
2. LQ-PID Control
Briefly, we introduce the LQ-PID control method proposed by Suh and Yang [14] , in which the optimal linear feedback control law is matched to the conventional PID control for the second order system by augmenting the integral of the output variable as a new state.
Consider the following second order model:
PPT Slide
Lager Image
where y ( t ) , u ( t ) , ζ and ωn are the output variable, the control variable, the damping ratio and the natural frequency. The initial conditions of y ( t ) and dy ( t ) / dt are specified.
The new augmented state variables are:
PPT Slide
Lager Image
The state-space representation of the augmented system becomes:
PPT Slide
Lager Image
PPT Slide
Lager Image
where
PPT Slide
Lager Image
PPT Slide
Lager Image
If we apply the different input gain according to and as
PPT Slide
Lager Image
where Ac is the closed loop system matrix.
The time delay system can be represented as the system without the time delay as where [23] .
The quadratic performance criterion is considered for the LQR formulation:
PPT Slide
Lager Image
with the assumption that the weighting factor, Q , is positive semi-definite and symmetric Q = QT ≥ 0 , and ρ is a positive value.
The linear feedback control law to regulate the state, x ( t ) , is obtained as follows:
PPT Slide
Lager Image
PPT Slide
Lager Image
where the optimal gain matrix, K = KT , is obtained by solving the ARE:
PPT Slide
Lager Image
Let the components of K and G be:
PPT Slide
Lager Image
PPT Slide
Lager Image
which we will hereafter refer to as the optimal Kalman gains of the LQ-PID.
The optimal control law of Eq. (9) is represented as the following PID control formula:
PPT Slide
Lager Image
The state feedback system with the control input, u ( t ) , of Eq. (8) is obtained by substituting Eq. (12) into Eq. (3):
PPT Slide
Lager Image
So, the closed loop system matrix, Ac , is:
PPT Slide
Lager Image
and its characteristic equation becomes:
PPT Slide
Lager Image
The closed loop transfer function of the second order system with the conventional PID controller can be expressed as follows:
PPT Slide
Lager Image
where
PPT Slide
Lager Image
and the transfer function of Eq. (1) is
PPT Slide
Lager Image
The characteristic equation of Eq. (19) is
PPT Slide
Lager Image
Fig. 1 shows the block diagram of LQR of the state feedback system with the augmented state variable, forming PID structure where the optimal tuning of the conventional PID controller is achieved by LQR.
PPT Slide
Lager Image
Structure of LQ-PID control
By corresponding Eq. (18) to Eq. (16), the parameters of the PID controller are matched with the optimal Kalman gains of the LQ-PID structure from Eq. (13).
It has the following relationship
PPT Slide
Lager Image
such that the PID controller is tuned by selecting the weighting matrices, Q and ρ , because the parameters of the PID controller are related to the solution, K , of the ARE through Eqs. (19) and (12), and ωn is given by the system plant.
Let the weighting matrix, Q , be as Eq. (20):
PPT Slide
Lager Image
where N = [ n 0 n 1 n 2 ] is assumed as a partitioned matrix of the weighting matrix, Q .
3. Design Methodology
In this section we describe the proposed design procedure of the LQ-PID controller to meet both the frequency and time domain specifications.
Given a specific set of design requirements in the time domain such as overshoot, and rise and settling times, the feasible area of design parameters will be determined by pole matching to the normalized simple third order system [15 , 18] . The parameters are then connected to weighting factors, Q and ρ , by the ARE. For the frequency specifications, we consider the limiting behaviour for the loop shaping method in which Kalman’s inequality is applied [16 , 21] . The time- and frequency-domain approaches are combined by formulating the optimal problem with constraints in terms of Q and ρ to satisfy both time- and frequency-domain specifications. We consider the cost function of the proposed optimal problem as the weighted summation with the 2-norm of control gains and the integral of the absolute error in order to avoid the saturation of input energy and reduce the overall effect of the error.
- 3.1 Combined constrains for time and frequency domain
- 3.1.1 Frequency domain performances
In the frequency domain, there are many performance indices such as gain margin, phase margin, infinite norm of the sensitivity function, and the complementary sensitivity function. The LQ-PID control guarantees the robust stability of the optimal system, which is verified by Kalman’s inequality [15] . The inequality verifies that the sensitivity performance is improved at all frequencies,
|S( )| ≤ 1 , for all ω [3, 4, 17, 19].
The sensitivity function is defined in the usual way,
PPT Slide
Lager Image
where the loop transfer function gLQ ( s ) of LQR is obtained as follows:
PPT Slide
Lager Image
In order to associate the frequency loop shaping procedure with the weighting factors, Q and ρ , we use Kalman’s identity for the single input single output case:
PPT Slide
Lager Image
where gOL ( ) = N ( jωI A ) −1 B with a partitioned matrix, N , of the weighting matrix, Q ,. Eq. (23) can be represented as:
PPT Slide
Lager Image
Deriving the magnitude of the complex function in Eq. (24), we have:
PPT Slide
Lager Image
Eq. (25) can be approximated as Eq. (26) only for low and high frequencies since | gLQ ( )| >> 1 and | gLQ ( )| << 1 , respectively [15] .
PPT Slide
Lager Image
The magnitudes of the loop transfer function, gLQ ( s ) , are approximated via the limiting behaviours of gLQ ( ) at low and high frequencies [19] , respectively, as:
PPT Slide
Lager Image
and
PPT Slide
Lager Image
Let each asymptotic line of | gLQ ( )| at the low and high frequency be lL ( ω ) and lH ( ω ) , respectively.
We consider a design procedure that involves explicitly shaping the magnitude of the loop transfer function, gLQ ( s ) , for frequency domain performances.
In classical loop shaping, better performances such as good command following and disturbance rejection are obtained when the magnitude of the loop transfer function, gLQ ( s ) , is larger for low frequencies at | gLQ ( )| >> 1 [19] . Fortunately, robust stability, noise attenuation, and control energy reduction are valid primarily in high frequencies at | gLQ ( )| << 1 . The shape of | gLQ ( )| then should not invade either the low and high frequency barriers, Q and β ( ω ) , respectively, as described in Fig. 2 .
PPT Slide
Lager Image
Typical shape of |gLQ( )| with barriers for the requirements
For this adequate loop shaping, we can develop a loop shaping technique in which the design parameter, N , plays a key role in maintaining | gLQ ( )| beyond the required boundaries. Specifically, this means that we match the weighting matrix, Q , of the LQ problem directly with the performances of the frequency domain to satisfy the boundary conditions via the approximation of limiting behavior.
Let Ω r be a boundary frequency of the command following barrier, α ( ω ) , and scalar r m be a magnitude of α ( ω ) , as shown in Fig. 2 . Set the barrier, α ( ω ) , as:
PPT Slide
Lager Image
Since lL ( ω ) should exist on the upper side of the barrier, α ( ω ) , the asymptotic line of | gLQ ( )| at the low frequency must satisfy:
PPT Slide
Lager Image
Solving Eq. (30) for the range of the parameters, n 0 and n 1 , gives:
PPT Slide
Lager Image
Let Ω r be a boundary frequency of the senor noise, and
PPT Slide
Lager Image
be the inverse of the maximum value of the modeling error in Fig. 2 .
The barrier, β ( ω ) , can then be expressed as:
PPT Slide
Lager Image
The inequality constraint of lH ( ω ) for high frequencies is derived from Eq. (28) in the same manner:
PPT Slide
Lager Image
Hence this constraint can be rearranged with the respect to the element n 2 of the partial matrix, N of Q as
PPT Slide
Lager Image
Finally, we can yield the constraints of Eqs. (31) and (34) with respect to the weighting factors, Q and ρ , of the LQ problem to satisfy frequency domain performances, adjusting the loop shaping of | gLQ ( )| .
- 3.1.2 Time domain performances
In the time domain, the desired transient performances are typically characterized by overshoot, and rise, and settling times. Designs for controllers that improve these transient performances of a second order closed loop system in the time domain are well explained in current literatures; however higher order systems are generally analyzed via the dominant pole concept as the reduced second order system is hard to formulate [24] .
In this paper, we consider the special formulation of the third order closed loop system as the target function which is referred to [14] as:
PPT Slide
Lager Image
The shape of the unit step response of Eq. (35) is determined by parameters p , r , and ωt . Furthermore, it is found that the overshoot is maintained constantly, regardless of the variations in natural frequency, ωt , when parameters p and r are fixed. This is verified from the formulation of the output, y ( t ) [25] .
PPT Slide
Lager Image
where
PPT Slide
Lager Image
Since the time, t , and the natural frequency, ωt , are united, we can modify the time, t , to a new time, τ , using ωt as the time ratio, as follows:
PPT Slide
Lager Image
where τ = * t ωt
The output y ( τ ) is represented as the response of the normalized third-order closed-loop system (NTCS), described as Eq. (35) with ωt = 1 . Hence the time ratio of the same step response y ( τ ) can be regulated independently on variations of the natural frequency, ωt as shown in Fig. 3 .
PPT Slide
Lager Image
Step response of the third order system according to variations in ωt
We consider two steps to formulate the constraints of time domain specifications. First, we determine the constraints with respect to p , r , and ωt to satisfy the performance requirement in the time domain. For the overshoot requirement, the constraint function is illustrated as follows:
PPT Slide
Lager Image
where fOS ( r, p ) represents the maximum step response function consistent with the variables, p and r .
Fig. 4 shows that the contours of maximum percentage overshoot with respect to the step reference can be plotted in the coefficient plane of p and r . If the Routh-Hurwitz stability criterion is applied to the normalized system transfer function of Eq. (35), Eq. (40) is required for simple absolute stability with h = 1 :
PPT Slide
Lager Image
PPT Slide
Lager Image
Coefficient plane for overshoot in NTCS
A gain margin of 0.33, which is allowable in practice, corresponds to the line of Eq. (40) for h = 1.5 [18] . The region between the two contours at h = 1 and h = 1.5 is the unallowable design area, which is distinct by oblique in the coefficient plane of Fig. 4 .
Hence, except in terms of area, the overshoot coefficient is settled according to the boundary of the required overshoot percentage. The natural frequency, ωt , can be addressed as the design parameter, which plays a role in regulating the rise and settling time under the overshoot determined by p and r .
PPT Slide
Lager Image
where ωST and ωRT depend on the values of the natural frequency, ωt , to satisfy the required settling and rising time, respectively. In other words, the system parameters p and r are determined to meet the given overshoot requirement, and ωt is considered as a factor used to separately satisfy the given settling time and rising time requirements.
Secondly, the system parameters p , r , and ωt should be formulated as a function with respect to the optimal Kalman gains of the LQ-PID feedback system via pole matching to equalize the characteristic equation of the target function Eq. (16) of NTCS.
PPT Slide
Lager Image
Given the above Eq. (42), the following three equations are derived.
PPT Slide
Lager Image
Using Eq. (43), weighting matrices, Q , and ρ , can be expressed in terms of the variables of the target transfer function via the ARE to satisfy the time domain performances, as follows:
PPT Slide
Lager Image
By substituting Eq. (43) to Eq. (44), the components [ n 0 n 1 n 2 ] of weighting matrix, Q are expressed as Eq. (45) in terms with the variables, p , r , and ωt for time domain performances, given input weighting, ρ , and system parameters, ωn , ζ and c of the plant
PPT Slide
Lager Image
where f 1 and f 2 are the functions to meet time domain specifications in terms of the system variables. Eq. (39) and (41) then can be represented by the function of the optimal Kalman gains through Eqs. (42), (43), and (44) as follows:
PPT Slide
Lager Image
PPT Slide
Lager Image
Therefore we can determine the feasible constraints Eqs. (46) and (47) in terms of positive weighting matrices, Q and ρ , of the LQ problem to meet transient step response requirements such as overshoot, rise, and settling time in time domain.
- 3.2 Formulation of an optimization problem
In the above sections, we derived that the constraint conditions Eqs. (31), (34), (46), and (47) to satisfy the combined time- and frequency-domain requirements. The conditions are represented as inequality functions with respect to weighting matrices, Q and ρ , of LQ problem, obtained via the loop shaping method and the pole matching method by the ARE. In this subsection, we formulate the optimal problem to determine optimal values of the design parameter N to minimize the object function subject to the constraints combined by time- and frequencydomain performances. We choose the weighted combination of two objectives, the 2-norm of PID gain and IAE. This objective of the optimal problem can vary depending on the control purpose of the designer in practice.
The first objective term is represented as Eq. (48) in order to avoid a high gain problem when we take ρ << 1for the cheap control as shown by Eq. (27) [26] .
PPT Slide
Lager Image
The second objective therm is the IAE index as Eq. (50) for the time response to consider the error arising from the difference between a nominal plant and a closed loop target function Eq. (35):
PPT Slide
Lager Image
These two objectives are combined by weighting vector, γ , as follows:
PPT Slide
Lager Image
where the weighting vector can be chosen according to the importance of each objective.
In order to combine the time- and frequency-domain performances, the optimization problem is formulated with respect to the object functions in Eq. (50) and the constraints in Eqs. (31), (34), (46), and (47), as follows:
minimize (n0, n1, n2) subject to
PPT Slide
Lager Image
fos(ρ,n0,n1,n2)≤Γ, ∀τ fWT(ρ,n0,n1,n2)≥max(ωSTRT), fixed p, r n0,n1,n2>0 for positive definite
where ωn andζ are given by the process, and Ω r , mr , Ω n ,
PPT Slide
Lager Image
, Γ , ωST , ωRT , and ρ are given as the design specifications.
The design procedure is summarized as
Summary of Design Procedure
Step 1 : Set parameter values of a given plant : (ζ , ωn ) .
Step 2 : Specify the frequency domain constraints with respect to the required design specifications (m r , Ω r , Ω n ,
PPT Slide
Lager Image
via the loop shaping method.
Step 3 : Specify the time domain constraints with respect to the required design specifications (Γ, ωST , ωRT ) via target function matching.
Step 4: Perform the optimization problem of Eq. (51) with the specific value of ρ to determine the weighting factor, Q = NT N , subject to the constraints obtained by Step 2 and 3.
Step 5: Determine the optimal PID control gain, G , by solving the ARE with the system parameters and the weighting matrix Q and R = ρ I determined in Step 4.
Step 6: Finish the process to determine the adequate controller if all of required design specifications are satisfied, if not reset the constraints, beginning with Step 2.
4. Simulation Examples
Example 1 : Consider the second order system with a small delay, referred in [27] as
PPT Slide
Lager Image
For the optimization, the constraints and weighting vector γ are given as follows
PPT Slide
Lager Image
fOS (ρ, n0, n1, n2) ≤ 9% fWT (ρ, n0, n1, n2) ≥ωST (8 sec) γ = [1 0.5], ρ = 0.0001
The frequency and time domain constraints are chosen to avoid the loop shape from the frequency barriers Q and β ( ω ) , and to satisfy that the overshoot and settling time of the target system are less than 9% and 8 sec, respectively.
According to the design procedures, design parameter N is determined as N = [0.0099 0.0151 0.005] , and PID controller parameters are obtained as Kp = 1.5027, Ti = 0.9967 and 0.5005 Td = via the ARE.
In Table 1 , it is shown that the proposed method only yields the adequate loop shape required to avoid invading the frequency barrier, a good step response to meet the time response requirements of less than 9% overshoot and 8 sec settling time, as well as simultaneously minimizing the cost function. Figs. 5 and 6 present the comparisons of the simulation results for frequency and time responses, respectively.
System performance comparisons of time and frequency responses for example 1.
PPT Slide
Lager Image
System performance comparisons of time and frequency responses for example 1.
PPT Slide
Lager Image
Comparison of the frequency shape of the loop transfer function with various other methods
PPT Slide
Lager Image
Comparison of step response with various other methods for example 1
The results of the simulation are compared with other PID control methods by Yang [27] , Ho [28] , Grassi [29] , and Olof [30] . As shown in the Table 1 and Fig. 5 , the responses of the proposed and Olof methods satisfy the frequency-domain specifications. Among them, only the proposed method meets the required time-domain specifications described as settling time (8 sec) and overshoot (9%) in Fig. 6 .
Example 2 : A fired heater problem is taken from [31] as an example of practical process; this is one of the most common examples with long time delay for the practical process. The open loop dynamic of the system is described by the transfer function, Gl ( s ) , at low fuel gas feed rates:
PPT Slide
Lager Image
For the optimization, the constraints and weighting vector γ are given as follows.
PPT Slide
Lager Image
fOS (ρ, n0, n1, n2) ≤ 10% fWT (ρ, n0, n1, n2) ≥ωST (100 sec) γ = [1 0.5], ρ = 0.0001
The design parameter N and PID controller are obtained as
N = [0.196 3.08 5.33]×10-3 Kp = , 0.221 Ti = 0.02, and Td = 0.4567 respectively.
Table 2 shows the results of the proposed controller and those by Grassi [29] , Olof [30] , Ho [28] , and Huang [32] . With the exception of the Ho method, they satisfy the time domain specifications with respect to the percent overshoot (10%) and settling time (100 sec). On the other hand, the loop shapes at the low and high frequency are obtained only by the proposed method, and Ho’s. Only by the proposed method presents the allowable good step response for the time domain performance simultaneously satisfied in addition to the frequency response to guarantee a magnitude higher than 15 db at the low frequency and lower than -40 db at high frequency. Figs. 7 and 8 present a comparison of loop shapes in the frequency domain and step responses in the time domain, respectively.
System performance comparisons of time and frequency responses for example 2.
PPT Slide
Lager Image
System performance comparisons of time and frequency responses for example 2.
PPT Slide
Lager Image
Comparison of the frequency shape of the loop transfer function with other methods for example 2
PPT Slide
Lager Image
Comparison of the step response with other methods for example 2
5. Conclusion
This paper proposes a new tuning method to combine time- and frequency-domain requirements for an LQ-PID controller. The weighting matrices, Q and R, are formulated to improve limiting behaviour of the loop transfer function by Kalman’s equality and target function matching is used for frequency and time domain performances, respectively. Both of them are simultaneously considered as the constraints of the optimal problem with the cost function in which the 2-norm of optimal control law and IAE index are applied. The PID control coefficients are obtained directly through the ARE, including the weighting factors Q and R. Simulation results have been presented to show the effectiveness of the proposed method in comparison to others. In conclusion, the proposed method guarantees not only the robust stability and design simplicity inherent in LQ-PID optimal control, but also considers the time- and frequency-domain performances for the second order system.
Acknowledgements
This work was supported by the Korea Institute of Energy Technology Evaluation and Planning (KETEP) grant funded by the Korea government Ministry of Knowledge Economy (No.20132010101870)
BIO
Chang-Hyun Kim He received his B.A. in Electronics Engineering from Kangnam University, Korea, in 2003. And he received his M.S and Ph.D in Electrical Engineering from Hanyang University, Korea, in 2006, and 2015, respectively. He was a visiting professsor at Kandahar University, Afghanistan from 2006 to 2007. He is currently a researcher in the Research Institute of Industrial Science of Hanyang University, Korea. His current research interests include the robust control, MPC, optimal digital control, and its application to network congestion control, magnetic levitation systems.
Ju Lee He received his M.S. degree from Hanyang University, Seoul, South Korea, in 1988, and his Ph.D. from Kyusyu University, Japan in 1997, both in Electrical Engineering, He joined Hanyang University in September, 1997 and is currently a Professor of the Division of Electrical and Biomedical Engineering. His main research interests include electric machinery and its drives, electromagnetic field analysis, new transformation systems such as hybrid electric vehicles (HEV), and high-speed electric trains and standardization. He is a member of the IEEE Industry Applications Society, Magnetics Society, and Power Electronics Society.
Hyung-Woo Lee He received the B.S. and M.S. degrees from Hanyang University in 1998 and 2000, respectively, and the Ph.D. degree from Texas A&M University, College Station, TX, in 2003, all in electrical engineering. In 2004, he was a Post-doctoral Research Assistant in the Department of Theoretical and Applied Mechanics, Cornell University. In 2005, he was a contract Professor at the BK division of Hanyang University. In 2006, he has been a Senior Researcher at the Korea Railroad Research Institute. He joined Korea National University of Transportation since 2013 and is currently an Assistant professor of the Dept. of Railway Vehicle System Engineering. His research interests include design, analysis and control of motor/generator, power conversion systems, application of motor drives such as Maglev trains, robots, and modern renewable energy systems.
References
Bellman R. , Dreyfus S. 1962 “Applied dynamic programming” Princeton University Press NJ
Pontryagin L. , Boltyanskii V. , Gamkrelidze R. , Mishchenko E. 1962 “The mathematical theory of optimal processes” Wiley Press
Anderson B. , Moore J. 1989 “Optimal control: Linear quadratic methods” Prentice Hall Press
Dorato P. , Cerone V. 1995 “Linear-quadratic control: An introduction” Prentice Hall Press
Shih Y. , Chen C. 1974 “On the weighting factors of the quadratic criterion in optimal control,” Int. J. Control 19 947 - 955    DOI : 10.1080/00207177408932688
Nekoui M. , Bozorgi H. 2012 “Weighting matrix selection method for lqr design based on a multiobjective evolutionary algorithm,” Advanced Materials Research 383 1047 - 1054
Ghoreishi S. , Nekoui M. 2012 “Optimal weighting matrices design for LQR controller based on genetic algorithm and PSO,” Advanced Materials Research 433 7546 - 7553
Kumar E. V. , Jerome J. , Srikanth K. 2014 “Algebraic approach for selecting the weighting matrices of linear quadratic regulator,” Green Computing Communication and Electrical Engineering (ICGCCEE) 1 - 6
Das S. , Pan I. , Halder K. , Das S. , Gupta A. 2013 “LQR based improved discrete PID controller design via optimum selection of weighting matrices using fractional order integral performance index,” Applied Mathematical Modelling 37 4253 - 4268    DOI : 10.1016/j.apm.2012.09.022
Moore J. , Mingori D. 1987 “Robust Frequency-shaped LQ Control,” Automatica 23 641 - 646    DOI : 10.1016/0005-1098(87)90060-4
Shi Y. , Becker T. C. , Furukawa S. , Sato E. , Nakashima M. 2014 “LQR control with frequency-dependent scheduled gain for a semi-active floor isolation system,” Earthquake Engineering & Structural Dynamics 43 1265 - 1284    DOI : 10.1002/eqe.2352
Silva D. L. , Paula C. F. , Ferreira L. H. 2013 “On the target feedback loop parameterization for H∞/LTR control,” Industrial Electronics (ISIE), 2013 IEEE International Symposium on 1 - 5
Stein G. , Athans M. 1987 “The LQG / LTR procedure for multivariable feedback control design,” IEEE Trans. Automat. Contr. 32 105 - 114    DOI : 10.1109/TAC.1987.1104550
Suh B. , Yang J. 2005 “A Tuning of PID Regulators via LQR Approach,” J. Chem. Eng. Japan 38 344 - 356    DOI : 10.1252/jcej.38.344
Yang J. , Suh B. 2007 “A new loop-shaping procedure for the tuning LQ-PID regulator,” J. Chem. Eng. Japan 40 575 - 589    DOI : 10.1252/jcej.40.575
Thomson M. , Cassidy P. , Sandoz D. 1989 “Automatic tuning of PID controllers using a combined time-and frequency-domain method,” Trans Inst. MC 11 40 - 47    DOI : 10.1177/014233128901100105
Burl J. 1998 “Linear optimal control: H2 and H∞ methods,” Addison Wesley Press
Kalman R. 1964 “When is a linear control system optimal?,” J. Basic Eng. 86 51 - 60    DOI : 10.1115/1.3653115
Athans M. 1986 “Lecture note on multivariable control system,” M. I. T. Lecture Note, Ref. No. 860224/6234
Towill D. 1966 “Analysis and synthesis of feedback compensatedthird-order control systems via the coefficient plane,” The Radio and Electronic Eng. 32 119 - 131    DOI : 10.1049/ree.1966.0064
Marques A. , Geerts Y. , Steyaert M. , Sansen W. 1998 “Settling time analysis of third order systems,” Electronics, Circuits and Systems, IEEE International Conference 505 - 508
Dorf R. C. , Bishop R. H. 2010 “Modern control systems,” 12th ed. Prentice Hall
He J. , Wang Q. , Lee T. 2000 “PI/PID controller tuning via LQR approach,” Chem. Eng. Sci. 55 2429 - 2439    DOI : 10.1016/S0009-2509(99)00512-6
Nise N. 2008 “Control systems engineering,” 5th ed. John Wiley & Sons Press
Ogata K. 2002 “Modern control engineering,” 4th ed. Prentice Hall Press
Kwakernaak H. , Sivan R. 1967 “Linear optimal control systems,” Wiley Interscience Press
Yang Y. , Wang Q. , Dai W. 2004 “Robust PID controller design for gain and phase margins,” J. Chem. Eng. Japan 35 874 - 879
Ho W. , Hang C. , Cao L. 1995 “Tuning of PID controllers based on gain and phase margin specifications’ Automatica 31 497 - 502    DOI : 10.1016/0005-1098(94)00130-B
Grassi E. , Tsakalis K. 1996 “PID controller tuning by frequency loop-shaping’ Proc. IEEE CDC Kobe, Japan 4776 - 4781
Olof G. 2009 “Design of robust pid controllers with constrained control signal activity,” Licentiate Thesis Lund University
Wang Y. , Rawlings J. 2004 “A new robust model predictive control method. II: examples,” J. Proc. Cont. 14 249 - 262    DOI : 10.1016/S0959-1524(02)00132-4
Huang H. , Lee M. , Chen C. 2000 “Inverse-based design for a modified PID controller,” J. Chin. Inst. Chem. Eng. 31 225 - 236