Implementation of a Fuzzy TSK Controller for a Flexible Joint Robot

Abstract

This paper proposes a fuzzy TSK controller to control a rotary flexible joint manipulator. The flexibility of joint is performed by means of a solenoid nonlinear spring, which is connected between actuator output and joint input in a bilateral connection form to transfer the produced torque; also the smooth model of frictions is used for modeling the dynamics of flexible manipulator. The effect of coulomb friction and also gearbox backlashes is decreased by a pulsation signal as an extra voltage that is added to the control voltage of actuator. Actuator dynamics is modeled by consideration of saturation mode of armature current. Experimental results demonstrate that the proposed controller has an ability to control flexible joint manipulator with a good performance.

1. Introduction

Robot systems with flexible joints are generally used in the aerospace industry, military, chemical, integrated circuits, and nuclear energy systems in order to perform commands with high sensitivity and accuracy [1]. It is demonstrated that the majority of robots have nonlinear dynamic and performance. Hence a high-accurate controlling requires special design and controllers [2]. Linear controllers are still used in robotics though in most cases this type of controllers may result in low performance responses and also encounters stability problems. Industrial environments, where there are strong electrical noises, have the worse conditions because electronic circuits and especially measuring systems are affected by power noises. Similar to other electromechanical systems, robots have uncertainties on their parameters model due to tolerance in physical quantities. In practice, model uncertainties and disturbance decrease the efficiency of designed controller and also limit the area of convergence. For years, efforts have been made to solve this problem and several methods such as robust, adaptive, and intelligent controllers were proposed [3].

Fuzzy controllers have been widely proposed and analyzed in the literature over the past decade. The most useful property of the fuzzy system is the ability to approximate the mapping from input to output with arbitrary accuracy. Especially, the Takagi-Sugeno-Kang fuzzy model [4] is an adequate fuzzy model for complex systems. This work proposes the design of a TSK-fuzzy controller, which assigns different linear control laws in order to control a flexible joint robot (FJR). These robots have high nonlinearity on dynamics and must be limited in the vicinity of equilibrium point. Such systems may be unstable in out of operating space because of high nonlinearity error. In these cases, even if designing a controller is being realizable, usually closed loop system has no good performance.

The proposed method has some advantages as follows:(1)improved performance controller for wide changing of payload,(2)no need for frequent calculations,(3)decreasing in controller time delay by eliminating some of the low effect sentences,(4)with adding a vibration signal we could reduce effects of fraction and backlash.

This paper subtends to four sections. In Section 2, we show how to make an acceptable model of Laboratory FJR (L-FJR), which has rigid arm, flexible joint, DC motor, and gearbox as actuator. Also in this section, a new compensator is proposed to decrease friction and backlash effects in FJR modeling. In Section 3, a controller designing based on fuzzy TSK is discussed. In Section 4, a number of methods are proposed for calculating, collecting, and measuring FJR parameters. This model will be completed and modified by applying parameters uncertainties. Also in this section, simulation results and experimental outputs are presented.

2. Modeling

The scheme of the single link FJR is shown Figure 1, which includes rigid arm, flexible joint, and actuator. Usually actuator is a combination of gearbox and electromagnetic motor as a compact unit fixed on the work-table by a rigid base. Several fixed ball-brings on parallel bases are used to fixing shafts and reducing effect of friction. Gearbox generated torque is transferred by fixed shafts from gearbox-out to input of spring and from output of spring to joint of link. Solenoid-type spring is selected for coupling [5]. In Section 2.4, it will be shown this spring is employed in torsion form and also has weak nonlinearity.

Figure 1 

Scheme of a single link FJR.

2.1. Base Model

Dynamic models for link and actuator sections can be created by two Lagrange differential equations as follows [6]: 𝑑𝑑𝑡𝛿𝐿_𝑖𝛿𝜃̇_𝑖−𝛿𝐿_𝑖𝛿𝜃_𝑖=𝑢_𝑖𝑖=1;linkofsystem,𝑖=2;actuatorofsystem, (2.1) where 𝐿_𝑖 represents difference between the kinetic and potential energy in link and actuator as follows: 𝐿_𝑖𝜃_𝑖,𝜃̇_𝑖=𝜅_𝑖𝜃_𝑖,𝜃̇_𝑖+𝑝_𝑖𝜃_𝑖, (2.2)𝜅_𝑖, 𝑝_𝑖 in link and actuator are expressed by the following equations: 𝑝link∶¹𝜃₁=12𝑚_𝐿⋅𝑔⋅ℓ⋅cos𝜃₁+𝑚_𝑃⋅𝑔⋅ℓ⋅cos𝜃₁,𝜅₁𝜃₁,𝜃̇₁=12𝐼_𝐿𝜃̇₁²+12𝑚_𝑃ℓ𝜃̇_𝑖²,𝑈₁=𝑈_𝐶−𝑈_𝐹𝐿−𝑈_𝐸1,𝑝actuator∶²𝜃₂𝜅=0,₂𝜃₂,𝜃̇₂=12𝐽⋅𝜃̇₂²,𝑈₂=𝑈_in−𝑈_𝐶−𝑈_𝐹𝑎−𝑈_𝐸2, (2.3)(2.4) where 𝑈_𝐸1, 𝑈_𝐸2 and 𝑈_𝐹𝑎, 𝑈_𝐹𝐿 are axles position shaft-encoders torques and friction torques of link and actuator. Also 𝑈_𝐶 and 𝑈_in are the transfer torques from actuator to link joint and gearbox out-torque. In these equations, 𝑚_𝐿, 𝑚_𝑃, 𝐼_𝐿, and 𝐽 are mass of link, payload, inertia moment of arm, and actuator, respectively. Also ℓ, 𝑔, 𝜃₁, and 𝜃₂ are the length of arm, the gravity acceleration constant, the rotation angle of link, and actuator-shaft to system base ({𝐴₀}), respectively. By replacing 𝐿₁ and 𝐿₂ in the Lagrange equations in (2.1), FJR dynamics is described by 𝐼_𝐿+𝑚_𝑃(ℓ)²𝜃̈₁+12𝑚_𝐿+𝑚_𝑃𝑔⋅ℓ⋅sin𝜃₁=𝑈_𝐶−𝑈_𝐹𝐿−𝑈_𝐸1,𝐽⋅𝜃̈₂+𝐵⋅𝜃̇₂=𝑈_in−𝑈_𝐶−𝑈_𝐹𝑎−𝑈_𝐸2. (2.5)

2.2. Friction Modeling

As it was mentioned 𝑈_𝐹𝑎, 𝑈_𝐹𝐿 are friction torques in the link and actuator which are generated by viscous and Coulomb frictions. Dynamic of friction torques can be written as [7] 𝑈_𝐹𝐿=𝐹_𝑉𝐿⋅𝜃̇₁+𝑈_𝐶𝐿,𝑈_𝐹𝑎=𝐹_𝑉𝑎⋅𝜃̇₂+𝑈_𝐶𝑎. (2.6)

Generally 𝐹_𝑉𝐿, 𝐹_𝑉𝑎 are constants which are related to type of surfaces and quality of lubrication. Usually, using lubrication and ball-brings can decrease Coulomb friction on the joint and arm sections. But lubrication cannot be used in brushing of DC motor, so Coulomb friction has always high value in gearbox and motor section. In this paper, 𝑈_𝐶𝑎 is selected as follows [8]: 𝑈_𝐶𝑎=𝐶_𝑓⋅sgn𝜃̇₂,𝜃̇₂1≠0,𝑊𝑈_in,𝑈_in≤𝑊⋅𝑆_𝑓𝑆_𝑓⋅sgn𝑈_in𝜃̇,otherwise,²=0, (2.7) where 𝐶_𝑓, 𝑆_𝑓, 𝑈_in, and 𝑊 are coefficient of kinetic and static frictions, applied torque and proportional weight respectively. It is also proved that Coulomb friction is generated by kinetic and static frictions and generally 𝐶_𝑓 is less than 𝑆_𝑓 (𝐶_𝑓<𝑆_𝑓). As an example, Figure 2 shows the modified figure of [9] for friction characteristic of a compact motor and gearbox.

Figure 2 

Friction characteristic of a compact motor and gearbox.

Often, designing a nonlinear controller for nonlinear systems will be conducted to use of numerical methods, so dynamic equations should have Taylor expansion. There is no Taylor series for equations similar to (2.7), because it has some nonderivativeable points on defined domain. This problem can be solved by using the following approximate equation (2.8) instead of (2.7). This equation is derivativeable in all defined domain 𝑈_𝐶𝑎=𝛽tanh𝐶⋅𝜃̇₂+𝐴⋅𝜃̇₂1+𝐵⋅𝜃̇²₂,𝜃̇₂∈ℜ. (2.8)

Also 𝑆_𝑓 and 𝐶_𝑓 of (2.7), which have positive values, are functions of 𝐴, 𝐵, 𝛽 as follows: 𝐶_𝑓𝑆=𝛽,_𝑓=Max𝛽tanh𝐶⋅𝜃̇₂+𝐴⋅𝜃̇₂1+𝐵⋅𝜃̇²₂, (2.9) where 𝐶 is a positive number which represents the slope of 𝑈_𝐶𝑎 in low speeds. Lubrication on ball-brings and gearbox is usually used in order to decrease 𝐶_𝑓 and 𝑆_𝑓. But lubrication is not possible between collector and coals in DC and universal motors, because it may cause electrical conductivity problem. Based on (2.7), Coulomb friction torque has high value at low speeds, so if rotor rotates without stopping, then starting torque is eliminated and Coulomb friction is decreased, which is shown experimentally in Section 4.

2.3. Motor Modeling

In most papers, the model of electromagnetic motor, which is used as driver, is described by a linear and low-order differential equation. In spite of simplicity, this model is used in most usual applications yet. But certainly linear model cannot be used in robots because the direction of rotor is always changed between CW and CCW status [10]. So nonlinearity and saturation must be considered in motor model especially in lightweight robots. Also in this paper, robot is actuated by a DC motor. Therefore, robot input torque is a function of armature current of DC motor as follows [10]: 𝑈_out-motor=𝐾_𝑚⋅𝑖_𝑎𝑖⋅tanh^𝑎𝑖_𝑎(Max), (2.10) where 𝑖_𝑎, 𝐾_𝑚, and 𝑖_𝑎(Max) are armature current, positive value as armature current to out-torque conversion constant, and maximum allowable current in armature, respectively. If the armature current is equal or greater than the maximum allowable current, saturation would happen in rotor and stator cores. In this paper, saturation is modeled by smooth function tanh(⋅) as an approximation of saturation dynamics. 𝑈_out-motor is applied to input of coupling spring as 𝑈_in. Also, armature current has a relation with the motor’s electrical parameters as follows [15]: 𝑖_𝑎=1𝑉𝐿𝑠+𝑅^𝑚−𝐾_𝑏𝜃̇_𝑚, (2.11) where 𝑅, 𝐿, 𝑉_𝑚, 𝜃̇_𝑚, and 𝐾_𝑏 are armature resistance, armature inductance, input voltage, motor speed at gearbox out, and speed to back e.m.f voltage conversion constant, respectively. Figure 3 shows a model of DC motor that is used as robot actuator. In this model, motor input voltage is a mixture of nonlinear controller output and symmetric pulse vibration voltages.

Figure 3 

Model of DC motor and input mixed voltage.

In order to actuate link in any position, it is necessary that motor out-torque must be always more than required torque.

2.4. Coupling Model

In FJRs, flexible gearbox out-torques is applied to robot joint as actuator. But in this paper, flexible gearbox was replaced by a set of rigid gearbox and tortuous solenoid spring for having high flexibility. In this working mode, spring coefficient is very low. Other problem, which complicates designing process, is nonlinearity of spring. As it is shown in Figure 4, effective diameter of spring can be changed in Tensional and Spastic modes, because distance of spring bases is always fixed.

Figure 4 

Tensional and spastic modes.

Increasing and decreasing in spring diameter on Tensional and Spastic modes can cause higher nonlinearities in FJRs specifications and also model may become asymmetric according to the following equations [11]: 𝑈_𝑐=𝑈⁺_𝑐=𝑎⁺₁𝜑+𝑎⁺₃𝜑³+𝑎⁺₅𝜑⁵𝑈,𝜑>0,_−𝑐=𝑎⁻₁𝜑+𝑎⁻₃𝜑³+𝑎⁻₅𝜑⁵𝑎,𝜑<0,_+𝑖≠𝑎⁻_𝑖>0or𝑎⁺_𝑖≠𝑎⁻_𝑖<0𝑖=1,3,and5, (2.12) where 𝜑 is defined as follows: 𝜑=𝜃₂−𝜃₁∣𝜃₁∈ℝ,𝜃₂∈ℝ. (2.13)

All 𝑎⁺_𝑖 and 𝑎⁻_𝑖 are Tensional and Spastic modes’ polynomial coefficients. In (2.12) it is assumed that coefficients of even powers of 𝑈⁺_𝑐 and 𝑈⁻_𝑐 polynomials are all zero, because of symmetric structure to origin. Generally (2.12) is not derivable on the origin; therefore, it will not have Taylor expansion. In order to eliminate conflicting, the following equation can be used instead of (2.12) in defined domain as a continues and derivable approximation 𝑈_𝑐=𝑎₁𝜑+𝑎₃𝜑³+𝑎₅𝜑⁵, (2.14) where 𝑎_𝑖 are new polynomial coefficients with uncertainty that are bounded between lower and upper size of 𝑎_𝑖. Therefore, 𝑎_𝑖 can be defined based on 𝑎⁺_𝑖 and 𝑎⁻_𝑖 as follows: 𝑎_𝑖=𝑎_𝑖1+𝑤_𝑎𝑖,𝑎_𝑖=𝑎⁻_𝑖+𝑎⁺_𝑖2∣𝑎⁻_𝑖&𝑎⁺_𝑖∈ℝ⁺or𝑎⁻_𝑖&𝑎⁺_𝑖∈ℝ⁻,𝑤_𝑎𝑖<𝑎⁻_𝑖−𝑎⁺_𝑖𝑎⁻_𝑖+𝑎⁺_𝑖, (2.15) where 𝑤_𝑎𝑖 represents maximum deviation 𝑎⁺_𝑖 and 𝑎⁻_𝑖 from 𝑎_𝑗. Figure 5 shows how to connect spring to bases. In this paper, a new method is proposed for symmetrical transferring of torque from actuator to spring where 50% of coupled torque is applied to the spring at 𝐴 by 𝑂𝐴 link and the rest is applied to 𝐴^′ by 𝑂𝐴^′ link and the fixed arc of spring (𝐴𝐴^′). Spring properties may change if connection points (𝐴,𝐴^′) are fixed by high-temperature methods such as soldering. Therefore, to avoid this problem spring and bases were fastened by screws and spanners in laboratory experiments. Also applied force on connection points (𝐴,𝐴^′) is equal to 𝑈⁻¹|𝑂𝐴|, which can be reduced by incrementing the bases length (𝑂𝐴, 𝑂𝐴^′). This structure has some advantages as follows:(1)symmetrical transmission for torque by two sides of spring bases,(2)decrease in applied forces on connection points,(3)decrease in friction forces of boll brings.

Figure 5 

Symmetric connection for spring and gearbox shaft.

3. Uncertainty and TSK Fuzzy Controller

In this section, the new model of FJR is presented, which has uncertainty as tolerance of parameters and disturbance inputs. It is also assumed that states of system are full available and system equations in state space are described as follows [12]: 𝑥̇=𝑓(𝑥)+𝑔₁(𝑥)𝑤+𝑔₂(𝑥)(𝑥)(𝑥)𝑢,𝑧=ℎ+𝑙𝑢, (3.1) where 𝑥∈𝑅^𝑛 is the state vector, 𝑤∈𝑅^𝑚1 is the disturbance input vector, 𝑢∈𝑅^𝑚2 is the control input vector, and 𝑧∈𝑅^𝑝 is the penalty variable vector. Also, it is assumed that functions 𝑓(𝑥), 𝑔₁(𝑥), 𝑔₂(𝑥), ℎ(𝑥), and 𝑙(𝑥) are defined and are smooth in the neighborhood of 𝑋 on 𝑅^𝑛.

3.1. Uncertainty

By using the discussions of previous section, new state space model of FJR can be written as follows: 𝑥̇₁=𝑥₂,𝑥̇₂=𝑓₂𝑥₁,𝑥₂,𝑥₃,𝑤,𝑥̇₃=𝑥₄,𝑥̇₄=𝑓₄𝑥₁,𝑥₃,𝑥₄,𝑥₅,𝑤,𝑥̇₅=𝑓₅𝑥₄,𝑥₅,𝑤+𝑔₂₅(𝑥)⋅𝑉_𝑚,𝑥̇₆=𝑥_𝑑−𝑥₁. (3.2)

These equations are written by using (2.5), (2.6), (2.10), (2.11), and (2.14), where 𝑤 is the input disturbance vector. The state vector on (3.2) is defined as: 𝑥₁=𝜃₁,𝑥₃=𝜃₂,𝑥₅=𝑖_𝑎,𝑥₂=𝜃̇₁,𝑥₄=𝜃̇₂,𝑥₆=𝑒⋅𝑑𝑡. (3.3)

Also the models of shaft encoders were eliminated for simplicity, because of their low effects. Equation for 𝑥̇₆ is added to FJR describing equations to decrease the tracking error. Where 𝑥_𝑑 and 𝑥₆ are the input command as desired position and integral of tracking error. In this paper, tolerances of mechanical parameters such as mass and length of link are also considered, because these parameters can be changed easily. It is assumed that measuring of fixed mechanical and electrical parameters in compact motor and gearbox has been done with enough precision. Tolerance of parameters can be defined as follows: 𝑚_𝐿=𝑚₁1+𝛼₁𝑤₁,𝑚_𝑃=𝑚₂1+𝛼₂𝑤₂,ℓ=ℓ₁1+𝛼₃𝑤₃, (3.4) where, 𝛼₁, 𝛼₂, and 𝛼₃ are positive numbers as maximum deviations of 𝑚_𝐿, 𝑚_𝑃, and ℓ respectively. Also 𝑤₁, 𝑤₂, and 𝑤₃ are external input signals, which are limited by |𝑤_𝑖|<1. Therefore, 𝑤 is defined as follows: 𝑤=𝑤₁,𝑤₂,𝑤₃,𝑤_𝑎1,𝑤_𝑎3,𝑤_𝑎5,𝑥_𝑑^𝑇. (3.5)

3.2. TSK Controller

T-S fuzzy models are universal approximations [13, 14]. The form of a nonlinear system is 𝑥̇=𝐹(𝑥)+𝑏(𝑥)𝑢, (3.6) where 𝑥=[𝑥₁,𝑥₂,…,𝑥_𝑛]^𝑇 is the state vector. 𝐹(𝑥) and 𝑏(𝑥) are the nonlinear functions of the plant and 𝑢 is the control signal 𝑢(𝑡)=[𝑢₁(𝑡),𝑢₂(𝑡),…,𝑢_𝑝(𝑡)]^𝑇. The 𝑖th (𝑖=1,…,𝑟) IF-THEN rule in the fuzzy rule are in the following form:

𝑅_𝑖: If 𝑍₁(𝑡) is 𝑀_𝑖1 and … and 𝑍_𝑔(𝑡) is 𝑀_𝑖𝑔𝑥̇(𝑡)Then=𝐴_𝑖𝑥(𝑡)+𝐵_𝑖𝑢(𝑡),𝑦(𝑡)=𝐶_𝑖𝑥(𝑡). (3.7) In the 𝑖th implication, 𝑟 is the number of fuzzy rules, 𝑀_𝑗 are the fuzzy sets, 𝐴_𝑖 and 𝐵_𝑖 are the 𝑛×𝑛 and 𝑛×𝑝 constant state matrixes, and 𝑥(0) is the initial state vector: 𝑥̇(𝑡)=∑^𝑟_𝑖=1𝑤_𝑖[𝑍(𝑡)]𝐴_𝑖𝑥(𝑡)+𝐵_𝑖𝑢(𝑡)∑^𝑟_𝑖=1𝑤_𝑖[𝑍(𝑡,𝜇)]_𝑖[𝑍(𝑡=𝑤)]^𝑖[𝑍(𝑡)]∑^𝑟_𝑖=1𝑤_𝑖[𝑍(𝑡⇒𝑥̇(𝑡)=)]^𝑟_𝑖=1𝜇_𝑖[𝑍(𝑡)]𝐴_𝑖𝑥(𝑡)+𝐵_𝑖𝑢(𝑡), (3.8)𝑤_𝑖 is its corresponding membership function, 𝑤_𝑖[𝑍(𝑡)]=∏^𝑔_𝑗=1𝐹_𝑗[𝑍_𝑗(𝑡)], and 𝐹_𝑗[𝑍_𝑗(𝑡)] is the grade of membership of 𝑍_𝑗(𝑡) in 𝐹_𝑗 (fuzzy sets).

The final state of fuzzy systems and the output is represented as 𝑥̇(𝑡)=^𝑟_𝑖=1𝜇_𝑖𝐴_𝑖𝑥(𝑡)+^𝑟_𝑖=1𝜇_𝑖𝐵_𝑖𝑢(𝑡),𝑦(𝑡)=^𝑟_𝑖=1𝜇_𝑖𝐶_𝑖𝑥(𝑡). (3.9) Considering (3.7), design of the controller via parallel distributed pole placement results in another fuzzy system with 𝑟 rules in the form of the following:

if 𝑍₁(𝑡) is 𝑀_𝑖1 and 𝑍₂(𝑡) is 𝑀_𝑖2 and … and 𝑍_𝑔(𝑡) is 𝑀_𝑖𝑔Then𝑢(𝑡)=−𝐾_𝑖𝑥(𝑡),𝑖=1,2,…,𝑟. (3.10) Then the output of the fuzzy state feedback controller is 𝑢(𝑡)=−∑^𝑟_𝑖=1𝜇_𝑖[𝑍(𝑡)]𝐾_𝑖𝑥(𝑡), where the eigenvalues of 𝐴_𝑖−𝐵_𝑖𝐾_𝑗 are in the left half-plane.

By substation (3.10) in to (3.7), 𝑥̇(𝑡)=^𝑟𝑟_𝑖=1𝑗=1𝜇_𝑖𝜇_𝑗𝐴_𝑖−𝐵_𝑖𝐾_𝑗𝑥(𝑡). (3.11) The nonlinearity in system dynamic is 𝑧₁=cos𝑥₁, (3.12) T-S fuzzy modeling of FJR results in two rules as follows 𝑧₁(𝑡)=cos𝑥₁(𝑡)=²_𝑖=1𝑀_𝑖𝑧₁(𝑡)𝑏_𝑖𝑥₁(𝑡), (3.13) where the membership functions are 𝑀₁𝑧₁(𝑡=𝑧)¹(𝑡)−(2/𝜋)cos⁻¹𝑧₁(𝑡)(1−2/𝜋)cos⁻¹𝑧₁(𝑡),𝑧₁(𝑡)𝑀≠0,1,otherwise,₂𝑧₁(𝑡)=cos⁻¹𝑧₁(𝑡)−𝑧₁(𝑡)(1−2/𝜋)cos⁻¹𝑧₁(𝑡),𝑧₁(𝑡)≠0,0,otherwise, (3.14)(3.15)𝑏 is slope of sectors and defined as 𝑧₁(𝑡)(𝑡)2=𝑏𝑥,𝑏=1or𝜋,𝑏₁=1(Zero),𝑏₂=2𝜋(NotZero). (3.16) Membership functions of 𝑍₁ are titled as “Zero,” “Not Zero.” These membership functions are shown in Figure 6. Hence, the proposed two rules of T-S fuzzy modeling can be defined as follows.

Figure 6 

Membership function 𝑀1(𝑧1(𝑡)) and 𝑀2(𝑧1(𝑡)).

Rule 1 (If 𝑧(𝑡) is zero). Then 𝑥̇(𝑡)=𝐴₁𝑥(𝑡)+𝐵⋅𝑢_𝑐𝐴₁=−0100𝑚𝑔𝑙𝑏₁+𝑘𝐼0𝑘𝐼0𝑘0001𝐽𝑘0−𝐽−𝐵𝐽−𝐾_𝑏𝐾_𝑚000𝐾𝐽𝑅𝐵=_𝑚𝑅. (3.17)

Rule 2 (If 𝑧(𝑡) is not zero). Then 𝑥̇(𝑡)=𝐴₂𝑥(𝑡)+𝐵⋅𝑢_𝑐𝐴₂=−0100𝑚𝑔𝑙𝑏₂+𝑘𝐼0𝑘𝐼0𝑘0001𝐽𝑘0−𝐽−𝐵𝐽−𝐾_𝑏𝐾_𝑚𝐽𝑅. (3.18)

The state feedback controllers are obtained as follows: 𝐾₁=[],𝐾−1069.5−75.988574.8473.981₂=[−913.34−44.88618.2673.981]. (3.19) The final state feedback gain is 𝐾=𝜇₁[𝑍(𝑡)]𝐾₁+𝜇₂[𝑍(𝑡)]𝐾₂.

4. Experimental Results

In order to analyze the proposed controller performance, a single flexible joint robot test-bed was prepared in Islamic Azad University of Ahar. Figure 7 shows the photo of this robot. This FJR has two main sections, electromechanic and control sections. Electro-mechanic section is split to compact gear-boxed DC motor as actuator, solenoid spring as torque transformer, and plastic rigid link as arm. Also, in order to avoid undesired movements, main shaft from gearbox output to joint of arm is fastened by fixed parallel bases on main foundation.

Figure 7 

Single link laboratory FJR test bed.

Controller section has four high speed microcontrollers. Gearbox-shaft and link positions are measured by high precision shaft encoders as measured output signals. Shaft encoders signal wires are connected to two separable and independent microcontrollers as inputs. These outputs are wired to the interrupt inputs of microcontrollers. Outset, the pulses of shaft-encoders are decoded and then rotational angles are separately calculated by a special algorithm in microcontrollers.

In the next phase, decoded information is sent to main computer by RS232 protocol as serial communication and serial to USB converter. The main controller is a software controller programmed in Visual Basic. The controller on main computer gets gearbox-shaft and link positions by serial line and control law coefficients of a file that was already created by the other program in MATLAB. Special numerical code as DC motor voltage is created by use of control law and positions information in controller.

The controller on main computer gets gearbox-shaft and link positions by serial line and control law coefficients of a file that was already created by the other program in MATLAB. Special numerical code as DC motor voltage is created by use of control law and positions information in controller.

Then this code is sent to the fourth microcontroller by serial line. In order to create apposite commands for generating PWM form DC voltage as motor power supply, DC voltage is generated by four IGBTs as a controllable bridge and PWM method. According to the controller commands, produced voltage can be altered in range from −24 V to +24 V.

4.1. System Mechanical and Electrical Parameters

Table 1 shows values of link and spring mechanical parameters.

Table 1 

Link and spring properties.

Also, Table 2 shows values of compact gear-boxed DC motor set electrical and mechanical parameters. Figure 8 shows scheme of single link FJR and controller test-bed.

Table 2 

Compact DC motor and gearbox set properties.

Figure 8 

Scheme of single link FJR and controller test-bed.

Coulomb and Viscose friction coefficients are presented for link and compact gear-boxed DC motor in Table 3. This table also shows coefficients at two states: without vibration and with vibration signal. Of course vibration signal had proper quantities during experimentally tests.

Table 3 

Values of friction coefficients.

4.2. Compensated Model for Frictions

The friction model can be compensated by applying a pulsation form signal (vibration signal) with special frequency and amplitude. Tables 4 and 5 show values of 𝑆_𝑓, 𝐶_𝑓 by some vibration signals. These tables were gathered by various experimentally tests on test-bed FJR.

Table 4 

𝑆_𝑓 for some vibration signals.

Table 5 

𝐶_𝑓 for some vibration signals.

Figure 9 shows a test result for measuring of 𝑆_𝑓, 𝐶_𝑓. This test was done with square wave signal as a vibration signal with 𝑓=15 Hz and 𝑣_𝑃−𝑃=4 v.

Figure 9 

𝑆𝑓, 𝐶𝑓 measuring test.

4.3. Effect of Vibration Signal on Friction and Backlash

In order to show advantages of proposed controller with vibration signal, a number of tests were done on FJR for various input commands. These tests are repeated with commands 45′, 120′, and 150′.

Figure 10 shows that response to command 45′ with vibration signal is more accurate than without it and also shows that link angle was sat to input command at 𝑡=8 s by vibration signal, but it has ringing form on response for without vibration signal even after 𝑡=16 s. Figures 11 and 12 show response of FJR to commands 120′ and 150′ that are in unstable domain. It means that FJR has been stabled even in unstable domain by proposed controller. Certainly it is predictable that link response to these commands would have high settling time overshoot.

Figure 10 

Response to 𝑥𝑑=45′ with vibration signal.

Figure 11 

Response to 𝑥𝑑=120′ with vibration signal.

Figure 12 

Response to 𝑥𝑑=150′ with vibration signal.

4.4. Wide Changing in Payload

In order to show advantages of proposed controller law to traditional. The other tests were done on FJR by use of three different payloads.

Figures 13 and 14 show the test results of controller 𝑘(𝑥,𝑚_𝑝) and 𝑘(𝑥) in the same conditions. Similar to Section 4.3, optimal vibration signal has been applied in these tests.

Figure 13 

Effect of 𝑚𝑝 changing with 𝑘(𝑥,𝑚𝑝).

Figure 14 

Effect of changing with 𝑘(𝑥).

It can be concluded from Figures 13 and 14 that new controller has more compatibility with 𝑚_𝑝. Also, in addition to solving backlashes problem and decreasing of friction effect, the proposed controller has more accuracy and higher performance compared to traditional controllers.

4.5. Uncertainties and Disturbance

Robustness of the proposed controller is verified by several experimentally tests. The tests have been conducted by changing the link length. Figure 15 shows the step response of link position in controlled system by proposed control law for link uncertainties. Also Figure 16 shows the response of link position to step command 60′, where external disturbances are applied as impact forces to link.

Figure 15 

Response to 𝑥𝑑=60′ with link uncertainty.

Figure 16 

Response to 𝑥𝑑=60′ with external disturbance.

The results of tests show that the proposed controller can guarantee the system stability against parameter tolerances and external disturbances. The effect of 𝑄₆ on new controller is verified by some tests, the results are shown in Figure 17.

Figure 17 

The effect of 𝑄6.

Results of Figure 17 show operation of controlled system can be improved by fine adjusting of 𝑄₆. Inappropriate value for 𝑄₆ can drive system toward unstable mode. Motor current curves are shown in Figure 18. They are measured in two modes: with vibration signal and without vibration signal.

Figure 18 

Applied voltage to DC motor for 𝑥𝑑=120′.

Figure 18 shows, if motor control voltage is mixed with vibration signal, armature current will have smooth function with low variations after 𝑡=6 s. It is needed to mention that in all tests, position of link and gearbox shaft have been measured by use of shaft encoder and data acquisition board.

5. Conclusion

In this paper, Takagi-Sugeno fuzzy approach for modeling nonlinear systems is represented. Using this method, the nonlinear elements of the system can be replaced by linear subsystems and result can be achieved in a simpler model and less complicated equations. Also saturation problem in actuator and Coulomb friction were modeled by derivable and smooth models instead of classical methods in this paper. Moreover, experimental results of proposed controller on FJR showed that TSK fuzzy-based controller has an ability to stabilize and control FJR with acceptable performance. Moreover, it was shown that backlashes and vast starting torque problems in gearbox can be solved by use of vibration signal. Also proposed approach can be applied to any multilink flexible on stable domain with a good performance. Also, it can be expanded as an extension of the technique to the more challenging cases of systems with unstable dynamics with minimum performance, such as flexible link robots.