Received June 10, 2021, accepted June 25, 2021, date of publication June 29, 2021, date of current version July 20, 2021. Digital Object Identifier 10.1109/ACCESS.2021.3093427 Analytic Time Domain Specifications PID Controller Design for a Class of 2nd Order Linear Systems: A Genetic Algorithm Method CHIA-LING LEE AND CHAO-CHUNG PENG Department of Aeronautics and Astronautics, National Cheng Kung University, Tainan 701, Taiwan Corresponding author: Chao-Chung Peng (ccpeng@mail.ncku.edu.tw) This work was supported by the Ministry of Science and Technology under Grant MOST 107-2221-E-006-114-MY3 and Grant MOST 108-2923-E-006-005-MY3. ABSTRACT In this paper, an analytical approach of a Proportional-Integral-Derivative (PID) controller design for a servo motor position control with precise desired performance demands is presented. When designing linear controllers, the resulting closed-loop system is often considered as an approximation of a standard 2nd order system to simplify the gain design process. However, in reality, model discrepancies could exist in such approximation, which inevitably lead to performance mismatches between the actual system’s behavior and the desired one. In order to solve this issue, this paper presents an analytical PID controller solution, which is able to achieve desired time domain specifications precisely. In addition, a disturbance is added to the system and the associated PID controller will be designed to reject disturbance. Given a 2nd order linear system along with a PID controller, the related analytical solutions are derived first. Next, the associated fitness functions and constraints are constructed under given prescribed time domain specifications. Lastly, the PID control gains are determined using Genetic Algorithm. To demonstrate the effectiveness of the proposed solution, many numerical simulations are performed. A set of control gains is also found using the standard 2nd order system formulas in order to prove the accuracy of the presented method. Moreover, a comparison study with respect to the Matlab PID Tuner is considered to illustrate the advantage of the proposed PID design scheme. The main contributions of this study include precisely satisfying given control performance demands with exact analytical solutions, avoiding the existing model discrepancies between the standard 2nd order system and the exact closed-loop model. Simulations firmly prove that the proposed method can fulfill users’ different control demands as well as remove the frequently used trial-and-error strategy. INDEX TERMS PID analytical solution, genetic algorithm, PID tuning, servo motor control, time domain specifications. I. INTRODUCTION The well-known proportional-integral-derivative (PID) controller is one of the most commonly used control algorithms in linear/nonlinear systems [1] and in industrial fields [2]. It also plays a significant role in control systems related to training and education. PID controller has served as one of the major control methods for several engineering industries such as mechanical, electrical, chemical, aerospace, and so forth. The linear servo motor system is also a widely used The associate editor coordinating the review of this manuscript and approving it for publication was Nasim Ullah 99266 . candidate that the PID was integrated for speed as well as position control [3]–[5]. Several controllers consisting of extended concepts from the traditional PID controller have been proposed. These include fractional-order PID [6], [7], higher-order PID [8], integer-order PID [9], etc. However, in general, the traditional PID controller is still broadly used because of its low-cost implementation, feasibility, and robustness. Most importantly, the traditional PID controller is relatively easy for students and even engineers to comprehend as well as apply in real control systems. Therefore, the traditional PID configuration is considered and verified in this paper. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ VOLUME 9, 2021 C.-L. Lee, C.-C. Peng: Analytic Time Domain Specifications PID Controller Design The disturbance rejection, a very popular topic in control systems, is also discussed [10]. The purpose of disturbance rejection is to cope with expected and unexpected forces, which cause the system to drift away from its desired steady-state value. Nevertheless, the most commonly seen issue when using the PID controller is the selection of gain parameters. To deal with this problem, many classical control theories such as root-locus, phase lead/lag, Nyquist, and pole-placement [11] techniques can be used to assist with the gain determinations. The Ziegler-Nichols rule is also a popular method for PID tuning [12]. For practical implementations, most of the methods widely used in industrial processes are based on experience of field operators, actual experimental results, or both. Unfortunately, approximated time domain specifications such as overshoot, peak time, and settling time can roughly be achieved. As a result, fine tunings, which may be time consuming, are usually unavoidable. From the viewpoint of control performance evaluations, time response specifications are generally accepted as the most forthright way when observing behaviors of control applications. In order to achieve the desired performance specifications precisely when applying the traditional PID controller, in this study, the Genetic Algorithm (GA) is applied. GA is a search-based technique with the purpose of solving optimization problems containing linear and nonlinear constraints, of which include PID controller parameter tuning [13]–[15]. It has also been applied for disturbance rejection [16]. GA consists of several advantages such as its ability to avoid trapping in local optima and its wide solution space search. The main purpose of this work is to achieve precise time domain specifications for a class of 2nd order linear systems controlled by a PID controller. A servo motor model is considered as a plant for demonstration. An external disturbance is applied to the system. Related time domain analytical solutions are then derived accordingly. These solutions in addition to time domain specifications are further considered simultaneously to construct related fitness functions as well as constraints. Without any solution approximation, the GA is applied for the optimal PID gain determination. The main contributions of this paper can be highlighted as follows: (a) analytical solutions subject to time domain specifications when applying the PID controller for a class of 2nd order systems are derived; (b) constrained fitness functions are then constructed to meet the users’ demands without approximations; (c) a Genetic algorithm solver is applied to find the specific control gains, which meet time domain specifications precisely; (d) different from the existing classical control design methods, the proposed method is capable of realizing time domain specifications without the use of trial-and-error strategies. Finally, a comparison study with respect to the Matlab PID Tuner is considered to illustrate the advantage of the proposed PID tuning algorithm. VOLUME 9, 2021 II. SERVO MOTOR DESCRIPTION AND PARAMETER IDENTIFICATION A. SERVO MOTOR DESCRIPTION The mechanical equation of a servo motor is described as J θ̈ (t) + C θ̇ (t) = u (t) (1) where θ represents the angular position which is available via the encoder, J denotes the motor inertia and C is the viscous friction coefficient. Before the design of the PID controller, all parameters should be available in advance. Therefore, system parameter identification needs to be addressed. Furthermore, in order to simulate the real situation for servo motor control, only the angle position is available via an encoder. Therefore, the rotation speed ω (k) is estimated by applying the central difference derivative θ (k + 1) − θ (k − 1) (2) ω (k) = 2Ts where Ts is the sampling interval. B. PARAMETER IDENTIFICATION Consider the bilinear transform discretization 2 1 − z−1 s≈ (3) Ts 1 + z−1 Substituting (3) into (1) yields the following discrete-time differential equation CTs Ts ω− (k) = − ω+ (k) + u+ (k) (4) 2J 2J where ω− (k) = ω (k) − ω (k − 1) ω+ (k) = ω (k) + ω (k − 1) u+ (k) = u (k) + u (k − 1) (5) Afterwards, forming (4) as the following compact form ω+ (k) u+ (k) ω− (k) ω− (k + 1) ω+ (k + 1) u+ (k + 1) opt X1 = .. .. .. opt X2 . . . | {z } ω− (k + n) ω+ (k + n) u+ (k + n) {z } | {z } X2×1 | Y(n+1)×1 A(n+1)×2 (6) where Ts CTs opt , X2 = (7) 2J 2J To find the optimal estimates, apply the least squares solution −1 X opt = AT A AT Y (8) opt X1 =− Eventually, the parameters of servo motor are obtained by Ts opt , opt 2Ĵ X1 Ts (9) 2X2 The estimated parameters will be used in the PID control gain determination. Ĵ = Ĉ = − 99267 C.-L. Lee, C.-C. Peng: Analytic Time Domain Specifications PID Controller Design III. DESIGN OF PID CONTROLLER UNDER TIME DOMAIN SPECIFICATIONS A. CONTROLLER DESIGN Consider a standard feedback system consisting of a 2nd order plant Gp (s) and a controller Gc (s) with the reference command Rθ (s) as shown in FIGURE 1, where D (s) denotes the external disturbance. In terms of the definition of M , N , P, Q, T , U , V , Z and the detailed derivations, please refer to the Appendix. B. TIME DOMAIN SPECIFICATIONS As illustrated in FIGURE 2, time domain specifications for control system designs often contain certain characteristics related to the system time response. In this paper, the characteristics for a step response focus on overshoot Mp , peak time tp , and settling time ts . FIGURE 1. System structure. Define the plant as a servo motor, of which the transfer function (TF) is 1 (10) s (Js + C) The reference command as a step input r (11) Rθ (s) = s is considered in this work where r is the input size. Thus, the system error can be expressed as E(s) = Rθ (s) − θ (s). A disturbance as a step function d D (s) = e−τ s (12) s is applied where d is the input size. Note that τ denotes the time-delay coefficient. The TF of the PID controller Gc (s) is kI Gc (s) = kP + + kD s (13) s where kP is the proportional gain, kI represents the integral gain, and kD denotes the derivative gain. Hence, the closed-loop TF can be written as Gc Gp Gp θ (s) = Rθ (s) + D (s) (14) 1 + Gc Gp 1 + Gc Gp The output solution for reference command and disturbance will be derived separately. The output solution for the reference command is θ1 (t) = r − Peλ1 t − e−Mt [Q cos (Nt) + T sin (Nt)] (15) The output solution introduced by the disturbance is Ueλ1 t 0 + Ve−Mt 0 cos Nt 0 0 θ2 (t) = H t (16) 0 +Ze−Mt sin Nt 0 where t 0 = t − τ . H represents the Heaviside step function ( 0, t < τ H (t − τ ) = (17) 1, t ≥ τ Combining both solutions derived from reference command and disturbance yields the analytical solution θ (t) = θ1 (t) + θ2 (t) 99268 (18) FIGURE 2. Design of overshoot, peak time and settling time. The definition of overshoot Mp is the maximum value the response overshoots the steady-state value. Peak time tp is the time required for the response to reach the point of maximum overshoot. Due to the effects from disturbances, two different values of each specification will be considered. In other words, Mp1 , tp1 , ts1 and Mp2 , tp2 , ts2 represent what happens to the system before and after the disturbance, respectively. To express overshoot analytically, recall that Mp = θ tp − θss /θss × 100% (19) where θss is the steady-state value that can be calculated by θss = lim sθ (s) = κ (20) s→0 According to the definition, it is evident that overshoot occurs when θ̇ (t) is zero. Therefore, taking the derivative of (18) separately gives θ̇1 (t) = −λ1 Pe λ1 t −Mt +e (QM − NT ) cos (Nt) + (QN + TM ) sin (Nt) (21) θ̇2 (t) =H t 0 λ1 Ue λ1 t 0 +e −Mt 0 (NZ − MV ) cos Nt 0 − (NV + MZ ) sin Nt 0 (22) In (21) and (22), it has been assumed that the derivative of H is zero since the disturbance in (12) is time-invariant. Peak time tp1,2 is then obtained by satisfying θ̇1 tp1 = 0 & θ̇2 tp2 − τ = 0 (23) VOLUME 9, 2021 C.-L. Lee, C.-C. Peng: Analytic Time Domain Specifications PID Controller Design Substituting tp1,2 into (19) gives λ 1 tp −Mtp Q cos Ntp × 100% (24) Mp1 = −Pe −e +T sin Ntp λt Ue 1 p2 + Ve−Mtp2 cos Ntp2 × 100% Mp2 = (25) +Ze−Mtp2 sin Ntp2 Eq. (24)-(25) is the analytical solution for overshoot. Settling time ts is the time it takes for the time response curve to reach and stay within a particular range of the steady-state value of the system. θ (ts ) = θss × (100 ± α) % (26) where α represents different values of percentages that can be applied in the following derivations. IV. GENETIC ALGORITHM FORMULATION FOR PID PARAMETERS DESIGN Based on the derivations shown in Section III, it can be found that design of PID control gains to precisely fulfill the time domain specifications is not a simple task. To satisfy the nonlinear equations given the time domain specifications, the GA is applied for PID gain searching. The fitness function and constraints are formed to ensure the system fulfill users’ desired performance demands including peak time tp,d , overshoot Mp,d , and settling time ts,d . In the following segments, three different cases will be discussed separately with their corresponding fitness function and constraints. A. REFERENCE COMMAND WITHOUT DISTURBANCE In this case, the reference command in (11) is considered. The input size of disturbance is set to be 0. D (s) = 0 (27) The desired specifications are tp,d1 , Mp,d1 , ts,d1 . The fitness function can be expressed as θ1 (t) − θss (28) where t = ts,d1 + ρ. ρ is the tolerance of the time required for the system to reach its steady-state after the desired settling time (5% of ts,d1 is recommended). The associated constraints are shown as follows. θ̇1 tp,d1 = 0 (29) θ1 tp,d1 − Mp,d1 = 0 (30) θ1 ts,d1 − θss − α 100 = 0 (31) |θ1 (t) − θss | − α 100 < 0 (32) where (29) ensures that the desired peak time satisfies its definition expressed in (23), (30) is applied based on (24), (31) is set to fulfill the settling time requirement. Note that (32) must be guaranteed when t > ts,d1 to ensure the time response curve will indeed stay within the α% range. VOLUME 9, 2021 B. PURE DISTURBANCE REJECTION In this case, the problem turns into a regulation problem, which means the input size of the reference command is 0. Rθ (s) = 0 (33) The disturbance in (12) is applied. Note that the steady-state value θss is recalculated by θss = lim sθ (s) = 0 s→0 (34) The desired specifications are tp,d2 , Mp,d2 , ts,d2 . The time-delay coefficient τ is also considered. The fitness function can be expressed as θ2 (t) − θss where t = ts,d2 + ρ. The associated constraints are shown as follows. θ̇2 tp,d2 − τ = 0 θ2 tp,d2 − τ − Mp,d2 = 0 θ2 ts,d2 − τ − θss = α 100 |θ2 (t) − θss | − α 100 < 0 (35) (36) (37) (38) (39) where (36)-(38) share the same feature as (29)-(31). In (39), t > ts,d2 is applied. C. POSITIONING IN THE PRESENCE OF EXTERNAL DISTURBANCE This case is a combination of two previous cases. The reference command in (11) and the disturbance in (12) are considered. The specifications Mp,d1 , tp,d1 , ts,d1 and Mp,d2 , tp,d2 , ts,d2 are applied before and after the disturbance happens, respectively. The time-delay coefficient is τ . The fitness function can be expressed as θ1 (t) + θ2 (t − τ ) − θss (40) where t = ts,d1 + ρ. The associated constraints are shown as follows. θ̇1 tp,d1 = 0 & θ̇2 tp,d2 − τ = 0 (41) θ1 tp,d1 − Mp,d1 < 0 & θ2 tp,d2 − τ − Mp,d2 < 0 (42) ( |θ1 (t) − θss | − α 100 < 0, t = ts,d1 + ρ1 θ1 ts,d1 + θ2 (t) − θss − α 100 < 0, t = ts,d2 + ρ2 (43) In (43), the tolerance ρ1,2 guarantees that the time response curve will indeed stay within the α% range when t > ts,d2 . Note that the value of ρ1 has a constraint of ρ1 < τ − ts,d1 (44) to ensure it doesn’t conflict with the physical response after the appearance of disturbance. Since there are only three variables to meet six specifications, the constraints in this case are set differently from previous cases. The constraints in (42) serve to let the system 99269 C.-L. Lee, C.-C. Peng: Analytic Time Domain Specifications PID Controller Design have overshoots that are less than the desired overshoots, instead of being the exact same as previous cases. In addition, the constraints for settling time are not as strict. The equation that ensures the settling time of the system will be the exact desired settling time, such as (31) and (38), has been removed. In terms of upper bound and lower bound in GA, it is important to choose the correct range. Incorrect selection will decrease the efficiency when searching for feasible solutions or even lead to infeasible solutions. Therefore, kP , kI , kD ∈ [0, ∞] is applied for all three cases in this paper to guarantee that the controller parameters are physically feasible. V. NUMERICAL SIMULATIONS In this section, numerical simulations for both with and without disturbances are performed to verify the work mentioned in previous sections. A flowchart is made and shown in FIGURE 3 for the purpose of demonstrating how the controller design process works. Step 3 (Problem Formulation Applying GA): Based on the demands in (48), the fitness function is θ1 (t) − θss (49) where t = ts,d + 0.05 × ts,d = 1.05. A 5% of ts,d is chosen for the tolerance ρ here. According to (48), the resulting constraints are expressed as θ̇1 (0.5) = 0 θ1 (0.5) − 12% = 0 |θ1 (1) − θss | − 0.02 = 0 |θ1 (t) − θss | − 0.02 < 0 (50) (51) (52) (53) where (53) is the associated constraint when t > 1. Regarding the PID gain searching, the Matlab Genetic Algorithm Toolbox is applied. Note that several optimization options are adjusted to ensure the accuracy of the optimal solution. These include setting constraint tolerance, max stall generations, max generations, and function tolerance. Their parameters are set to be 10−2 , 150, 500 and 10−3 , respectively. The parameters returned by the GA are shown as follows. (kP , kI , kD ) = (8.2588, 0.5760, 1.8459) (54) Therefore, the PID controller that meets the desired performance demands in (54) is FIGURE 3. Procedure for designing PID controllers using GA. Dc (s) = 8.2588 + Step 1 (Parameters Identification): Consider the system structure in FIGURE 1. The reference parameters J = 0.3 and C = 0.5 are chosen. Simulink in Matlab is used to generate measurement data for parameter identification. By following the procedure shown in Section II, a set of estimated parameters Ĵ = 0.3018 and Ĉ = 0.5045 are obtained. The plant for PID controller design is shown as follows. 1 G (s) = (45) s (0.3018s + 0.5045) 0.576 + 1.8459s s (55) Step 4 (Validate the Result With Numerical Simulations): The simulation displayed in FIGURE 4 illustrates the time response and its time domain specifications with the plant described in (54) and the controller form in (55). A. REFERENCE COMMAND WITHOUT DISTURBANCE Let the input size of the reference command be 1. 1 Rθ (s) = (46) s Note that the disturbance is not included in this case. That is, D(s) = 0 (47) Step 2 (Consider Desired Demands): The following desired performance demands are applied. tp,d = 0.5, Mp,d = 12%, ts,d = 1, α = 2% (48) The goal is to find a PID controller which makes the parameters kP , kI , kD meet the specifications in (48) precisely without any deviation. 99270 FIGURE 4. Numerical simulation applying the proposed GA based PID control parameter design scheme. It is obvious that the time response is able to reach the desired demands in (48) precisely. VOLUME 9, 2021 C.-L. Lee, C.-C. Peng: Analytic Time Domain Specifications PID Controller Design For comparison, the classical standard 2nd order formulas are also applied to show the influence caused by model discrepancies. Consider the general form of a standard 2nd system G2nd (s) = ωn2 s2 + 2ζ ωn + ωn2 (56) where ζ and ωn stand for damping ratio and natural frequency, respectively. The well-known standard 2nd order system formulas for time domain specifications including overshoot and settling time are provided in (57). ! 4 πζ , ts = Mp = exp − p (57) 2 ζ ω n 1−ζ To approximate a 3rd order system expressed in (14) as a 2nd order system, rewrite (14) as DPID (s) θ (s) = · G2nd (s) Rθ (s) s − λ01 FIGURE 5. Numerical simulation applying the standard 2nd order system formulas. (58) where DPID (s) consists of PID control gains, λ01 represents the root to be determined, and G2nd (s) denotes the remaining 2nd order transfer function as shown in (56). Applying the desired performance demands Mp,d and ts,d to (57) gives the corresponding ζ and ωn . Therefore, a standard 2nd order system which satisfies the desired demands is obtained. Afterwards, with the purpose of approximating the system behavior of (59) as a 2nd order system, the dominant pole approximation is applied. Rewrite (58) as a form with two roots θ(s) DPID (s) ωn2 = · 0 0 Rθ (s) s − λ1 s − λ2 s − λ03 (59) To allow λ02 and λ03 to become the dominant poles in this system, let λ01 satisfy the following requirement. λ01 = k · max Re λ02 , Re λ03 (60) where k ∈ R. The value of k decides how much alike this 3rd order system is compared to a 2nd order system. If k is too small, the system behavior won’t be close enough to a 2nd order system; on the other hand, if k is too big, it would lead to high gain control. For generality, k ∈ [10, 20] is recommended. In this simulation, k = 15 is chosen. Therefore, a set of kP0 , kI0 , kD0 = (112.0061, 617.2041, 13.9819) is obtained. The simulation result is shown in FIGURE 5. It is evident that even though the desired overshoot can be achieved, there exist unacceptable design mismatches for both peak time and settling time. This is because when a 3rd order system is approximated as a 2nd order system, only the denominator part is considered. However, when coping with real systems, the numerator also contains control gains that can affect system behaviors, leading to unexpected time response. By this comparison study, it firmly demonstrates that the proposed VOLUME 9, 2021 FIGURE 6. MATLAB PID Tuner Toolbox for comparison study. GA based PID control gain design is able to precisely achieve time domain specifications. Then, we further consider the comparison study with the Matlab PID Tuner from the view point of end-users. The interface of the PID Tuner provided by Matlab R2018b is shown in FIGURE 6, where FIGURE 6(a) displays the user interface and FIGURE 6(b) illustrates the performance tuning variables provided by the PID Tuner. The comparison of the simulation results from the Matlab PID Tuner and the proposed method is shown in FIGURE 7. FIGURE 7 shows that the simulation result obtained from the Matlab PID Tuner fails to meet the desired specifications simultaneously. This is because only two adjustable tuning tools are provided in the PID Tuner Toolbox. Moreover, 99271 C.-L. Lee, C.-C. Peng: Analytic Time Domain Specifications PID Controller Design FIGURE 7. Numerical simulation applying the MATLAB PID tuner. FIGURE 8. Numerical simulation applying the proposed GA based PID control parameter design scheme. while it is possible for users to achieve two of the desired specifications, the process requires a certain amount of time with trial-and-error. The features in the toolbox don’t allow users to acquire the time domain specifications instantly, which means the users have to adjust the parameters based on visual observation. Thus, not only is this method more timeconsuming, the result is also less precise compared to the proposed method. On the contrary, as illustrated in FIGURE 4, it shows that the desired time domain specifications can be simultaneously achieved without any trial-and-error. In order to show that the proposed method can be applied to meet various desired specifications, a set of parameters to fulfill aggressive response demands is chosen. tp,d = 0.5, Mp,d = 50%, ts,d = 3, α = 2% (61) Based on (61), the fitness function is θ1 (t) − θss (62) where t = ts,d + 0.05 × ts,d = 3.15. According to (61), the resulting constraints are expressed as θ̇1 (0.5) = 0 θ1 (0.5) − 50% = 0 |θ1 (3) − θss | − 0.02 = 0 |θ1 (t) − θss | − 0.02 < 0 (63) (64) (65) (66) where (66) is the associated constraint when t > 3. The parameters returned by the GA are shown as follows. (kP , kI , kD ) = (10.0891, 18.4405, 0.9522) (67) The simulation displayed in FIGURE 8 illustrates the time response and its time domain specifications with the plant described in (45) and the controller form in (67). All the simulation verifications demonstrate that the proposed method is able to successfully achieve users’ time domain demands precisely. The main advantage is that the time domain specifications are able to be satisfied simultaneously via the derived analytical solution, where the associated 99272 PID parameters are solved by the proposed fitness functions. Therefore, it saves quite a lot of time for trial-and-error tuning. B. PURE DISTURBANCE REJECTION The input size of the reference command is 0. Rθ (s) = 0 (68) Consider a step disturbance with the size of 1 and let the time-delay coefficient be 2. 1 (69) s Step 1 (Consider Desired Demands): The following desired performance demands are applied. D(s) = e−2s tp,d2 = 0.5 + 2, Mp,d2 = 5%, ts,d2 = 1 + 2 (70) The goal is to find a PID controller which makes the parameters kP , kI , kD meet the specification in (70) precisely without any approximation. Step 2 (Problem Formulation Applying GA): Based on the demands in (70), the fitness function is θ2 (t) − θss (71) where t = ts,d + 0.05 × ts,d − 2 = 3.05. A 5% of ts,d is chosen for the tolerance ρ here. According to (70), the resulting constraints are expressed as θ̇2 (0.5) = 0 θ2 (0.5) − 5% = 0 |θ2 (1)| = 0.02 |θ2 (t)| − 0.02 < 0 (72) (73) (74) (75) where (75) is the associated constraint when t > ts,d1 = 3. Similarly, the Matlab Genetic Algorithm Toolbox is applied here and the settings are the same as before. VOLUME 9, 2021 C.-L. Lee, C.-C. Peng: Analytic Time Domain Specifications PID Controller Design FIGURE 9. Numerical simulation applying GA. FIGURE 10. Numerical simulation applying GA. The parameters returned by the GA are shown below. (kP , kI , kD ) = (13.1783, 31.3647, 3.9490) (76) Therefore, the PID controller that meets the desired performance demands in (70) is 31.3647 + 3.9490s (77) s Step 3 (Validate the Result With Numerical Simulations): The simulation displayed in FIGURE 9 illustrates the time response and its time domain specifications with the plant described in (45) and the controller form in (77). It is obvious that the time response is able to reach the desired demands in (70) precisely. Dc (s) = 13.1783 + 1 (78) s Consider a step disturbance with the size of 1 and let the time-delay coefficient be 2. Rθ (s) = 1 (79) s Step 1 (Consider Desired Demands): The following desired performance demands are applied. D(s) = e−2s Mp,d2 = 5%, ts,d2 = 1.2 + 2 (80) The goal is to find a PID controller which makes the parameters kP , kI , kD meet the specification in (80) precisely without any approximation. Step 2 (Problem Formulation Applying GA): Based on the demands in (80), the fitness function is VOLUME 9, 2021 θ1 (0.3) − 12% < 0 & θ2 (0.5) − 5% < 0 (83) ( |θ1 (0.8) − θss | − 0.02 = 0 (84) |θ1 (0.8) + θ2 (3.3) − θss | − 0.02 = 0 ( |θ1 (t) − θss | − 0.02 < 0, t = 0.8 + δ1 (85) |θ1 (0.8) + θ2 (t) − θss | − 0.02 < 0, t = 3.3 + δ2 where (85) is the associated constraint when t > ts,d1 = 0.8. The parameters returned by the GA are shown as follows. (86) 21.2365 + 4.2656s (87) s Step 3 (Validate the Result With Numerical Simulations): The simulation displayed in FIGURE 10 illustrates the time response and its time domain specifications with the plant described in (45) and the controller form in (87). It is obvious that the time response is able to meet the desired demands in (80) precisely. Dc (s) = 19.6220 + VI. CONCLUSION Mp,d1 = 12%, ts,d1 = 1 θ1 (t) + θ2 (t − 2) − θss (82) Therefore, the PID controller that meets the desired performance demands in (80) is Let the input size of the reference command be 1. tp,d2 = 0.5 + 2, θ̇1 (0.3) = 0 & θ̇2 (0.5) = 0 (kP , kI , kD ) = (19.6220, 21.2365, 4.2656) C. POSITIONING IN THE PRESENCE OF EXTERNAL DISTURBANCES tp,d1 = 0.3, where t = ts,d + 0.05 × ts,d = 0.84 (5% of ts,d is chosen for the tolerance ρ here). According to (80), the resulting constraints are expressed as (81) In this paper, we present analytical solutions of PID control gain design for a class of 2nd order linear systems. The main contribution of this work is that the designed PID control gains are able to meet users’ demands precisely without time-exhausted trial-and-error. From the application point of view, the analytical solutions of PID control with the 2nd order servo motor plant and a step disturbance are derived. Moreover, the time domain specifications including peak time, overshoot, and settling time are given as performance 99273 C.-L. Lee, C.-C. Peng: Analytic Time Domain Specifications PID Controller Design demands. Three different scenarios including reference command without disturbance, pure disturbance rejection, reference command with disturbance are discussed individually. GA is applied by using the proposed analytical solutions as fitness function as well as constraints to determine the controller parameters. Finally, numerical simulations are performed to verify the effectiveness of the proposed method. The result shows that the designed PID controller can successfully reach the prescribed performance demands. Thus, the position behavior of the motor, with and without disturbance, including both transient and steady-state are assured. A comparison study between the proposed method and standard 2nd order system formulas as well as Matlab PID Tuner is also provided to highlight the advantages of the proposed PID design scheme. The result shows that the developed method is able to achieve users’ demands precisely without the need of trial-and-error. The proposed design procedures can be applied to a wide range of related control topics and is suitable for educational purposes of PID control system design. In future works, the artificial neural network may be considered for the design of PID parameters for nonlinear systems as well as suppression of varying external loads. The output solution regarding reference command is s2 + (C+kD ) 2 s J C Js + kP kI J s+ J (88) Examining the characteristic equation, kI (C + kD ) 2 kP s + s+ =0 (89) C (s) = s3 + J J J the roots are 1 10 λ1 = − b+G+ (90) 3a G " √ # 1 1 3 10 10 λ2,3 = − b+ − G+ ±j G− 3a 2 G 2 G (91) where the parameters used in (90) and (91) are listed as follows C + kD kP kI a = 1, b = , c= , d = , J J J v q u u 3 11 ± 121 − 4130 t G= , 2 10 = b2 − 3ac, 11 = 2b3 − 9abc + 27a2 d (92) Applying partial fraction expansion gives θ1 (s) = r P Qs + R − − s s − λ1 (s − λ2 ) (s − λ3 ) where P= 99274 C + λ1 J λ2 λ3 −1 λ1 − λ2 − λ3 + , λ1 R= λ2 λ3 P λ1 (94) Rewrite (93) as θ1 (s) = P Q (s + M ) s + (R − QM ) r (95) − − s s − λ1 (s + M )2 + N 2 of which !1/2 λ2 + λ3 (λ2 + λ3 )2 M = s− , N = λ2 λ3 − (96) 2 4 Taking the inverse Laplace of (95) gives θ1 (t) = r − Peλ1 t − e−Mt [Q cos (Nt) + T sin (Nt)] (93) (97) where R − QM (98) N Eq. (97) is the analytical solution for reference command. Similarly, the output solution regarding disturbance is T = θ2 (s) = Js3 d e−τ s + (C + kD ) s2 + kP s + kI Applying partial fraction expansion gives Vs + W U + 2 θ2 (s) = e−τ s s − λ1 s − (λ2 + λ3 ) s + λ2 λ3 APPENDIX r θ1 (s) = − s s3 + Q = 1 − P, (99) (100) of which λ2 λ3 −1 d λ1 − λ2 − λ3 + , U = J λ1 λ1 λ2 λ3 d W = U− λ1 J λ1 V = −U , (101) Rewrite (100) as U V (s + M ) s + (W − VM ) −τ s θ2 (s) = + e (102) s − λ1 (s + M )2 + N 2 Taking inverse Laplace yields Ueλ1 t 0 + Ve−Mt 0 cos Nt 0 0 0 θ2 t = H t 0 +Ze−Mt sin Nt 0 (103) where W − VM (104) N Eq. (103) is the analytical solution for disturbance. t 0 = t − τ, Z= REFERENCES [1] J. Zhang and L. Guo, ‘‘Theory and design of PID controller for nonlinear uncertain systems,’’ IEEE Control Syst. Lett., vol. 3, no. 3, pp. 643–648, Jul. 2019. [2] T. Samad, ‘‘A survey on industry impact and challenges thereof [technical activities],’’ IEEE Control Syst. Mag., vol. 37, no. 1, pp. 17–18, Feb. 2017. [3] M. A. Taut, G. Chindris, and D. Pitica, ‘‘PID algorithm used for DC motor control,’’ in Proc. IEEE 24th Int. Symp. Design Technol. Electron. Packag. (SIITME), Oct. 2018, pp. 365–372. [4] A. Rai, D. K. Das, and M. M. Lotha, ‘‘LabVIEW platform based real-time speed control of a DC servo motor with fuzzy-PI controller,’’ in Proc. Int. Conf. Electr., Electron. Comput. Eng. (UPCON), Nov. 2019, pp. 1–4. VOLUME 9, 2021 C.-L. Lee, C.-C. Peng: Analytic Time Domain Specifications PID Controller Design [5] C.-C. Peng and C.-L. Lee, ‘‘Performance demands based servo motor speed control: A genetic algorithm proportional-integral control parameters design,’’ in Proc. Int. Symp. Comput., Consum. Control (ISC), Nov. 2020, pp. 469–472. [6] B. Hekimoglu, ‘‘Optimal Tuning of fractional order PID controller for DC motor speed control Via chaotic atom search optimization algorithm,’’ IEEE Access, vol. 7, pp. 38100–38114, 2019. [7] A. Rastogi and P. Tiwari, ‘‘Optimal tuning of fractional order PID controller for DC motor speed control using particle swarm optimization,’’ Int. J. Soft Comput. Eng., vol. 3, no. 2, pp. 150–157, 2013. [8] M. Huba, D. Vrancic, and P. Bistak, ‘‘PID control with higher order derivative degrees for IPDT plant models,’’ IEEE Access, vol. 9, pp. 2478–2495, 2021. [9] G. Altintas and Y. Aydin, ‘‘A comparison on genetic algorithm based integer order and fractional order PID control of magnetic bearing system,’’ in Proc. IEEE Int. Conf. Mechatronics (ICM), Feb. 2017, pp. 20–24. [10] J. Han, ‘‘From PID to active disturbance rejection control,’’ IEEE Trans. Ind. Electron., vol. 56, no. 3, pp. 900–906, Mar. 2009. [11] U. M. Nath, C. Dey, and R. K. Mudi, ‘‘Desired characteristic equation based pid controller tuning for lag-dominating processes with real-time realization on level control system,’’ IEEE Control Syst. Lett., vol. 5, no. 4, pp. 1255–1260, Oct. 2021. [12] C. C. Hang, K. J. Astrom, and W. K. Ho, ‘‘Refinements of the Ziegler–Nichols tuning formula,’’ IEE Proc. D, Control Theory Appl., vol. 138, no. 2, pp. 111–118, Mar. 1991. [13] M. A. Ibrahim, A. K. Mahmood, and N. S. Sultan, ‘‘Optimal PID controller of a brushless DC motor using genetic algorithm,’’ Int. J. Power Syst., vol. 2088, no. 8694, p. 8694, 2019. [14] D. C. Meena and A. Devanshu, ‘‘Genetic algorithm tuned PID controller for process control,’’ in Proc. Int. Conf. Inventive Syst. Control (ICISC), Jan. 2017, pp. 1–6. [15] A. A. M. Zahir, S. S. N. Alhady, W. A. F. W. Othman, and M. F. Ahmad, ‘‘Genetic algorithm optimization of PID controller for brushed DC motor,’’ in Intelligent Manufacturing & Mechatronics. Singapore: Springer, 2018, pp. 427–437. [16] R. A. Krohling and J. P. Rey, ‘‘Design of optimal disturbance rejection PID controllers using genetic algorithms,’’ IEEE Trans. Evol. Comput., vol. 5, no. 1, pp. 78–82, Feb. 2001. VOLUME 9, 2021 CHIA-LING LEE was born in Taoyuan, Taiwan, in 1998. She is currently pursuing the B.S. degree with the Department of Aeronautics and Astronautics (DAA), National Cheng Kung University (NCKU), Tainan, Taiwan. She is a member of the Intelligent Embedded Control Laboratory, DAA, NCKU. Her research interests include system modeling and identification, analysis of dynamics and controls, autonomous systems, and flight mechanics. CHAO-CHUNG PENG was born in Kaohsiung, Taiwan, in 1980. He received the B.S. degree from the Department of Aeronautics and Astronautics (DAA), National Cheng Kung University (NCKU), Tainan, Taiwan, in 2003, and the Ph.D. degree, in 2009. From 2008 to 2009, he was a Research Assistant with the Department of Engineering, Leicester University, U.K. From 2010 to 2012, he was a Postdoctoral Fellow with the Department of Mechanical Engineering, NCKU. He was a Senior Engineer with the Embedded System Development Section, Measurement and Automation Department, ADLINK Technology, in 2012. From 2014 to 2016, he was with the Automation and Instrumentation System Development Section, Iron and Steel Research and Development Department, China Steel Corporation (CSC). Since 2016, he has been an Assistant Professor with the Department of Aeronautics and Astronautics, NCKU. He was promoted as an Associate Professor, in August 2020. His research interests include high-performance motion control and applications, unmanned vehicle design, advanced flight control system development, autonomous robotics, intelligence SLAM technology, system modeling, and diagnosis. He was awarded a membership in the Phi Tau Phi Scholastic Honor Society, in 2009, and the Excellent Young Engineering Professor Award by Chinese Society of Mechanical Engineers (CSME), in 2019. 99275