As BLDCMs have such good features as simple construction, high reliability, light electromagnetic pollution, and high power density, they are used extensively in servo systems and low-power drive systems [1, 2].
FSTPI (four-switch three phase inverter) as a reconfiguration topology for fault condition of traditional six switch inverter has been widely used [3-5]. But the uncontrollable phase current causes unsymmetrical voltage vector in FSTPI, and its waveform is much of distortion from rectangular . A new speed control method using the acceleration feed forward compensation is proposed to improve the speed response characteristic for a fourswitch three-phase BLDCM . The disturbance torque estimation method is adopted to improve the robustness of the method. But it is just verified to be economical and efficient in some occasions with light load, such as robot arm. A study on the generated torque ripples due to phase commutation is presented in the four-switch three-phase BLDCM , in which a current control technique is developed to minimize commutation torque for the entire speed range. But Hall sensor is used in this method and the cost of the system is raised. A novel direct torque control scheme  including the actual pre-stored back-EMF constants vs. electrical rotor position look-up table is proposed for BLDCM drive with two phase conduction scheme using four-switch inverter. And low-frequency torque ripples and torque response time are minimized. But the control method is more complicated, and does not discuss flux change characteristics.
In this paper, a novel voltage space vector modulation strategy and an optimal switching table are proposed to provide a basic condition for improve torque control. A single neuron adaptive PID method is proposed to observe flux for providing the accurate control information. The operating principle of FSTPI for the three-phase BLDCM drive and the proposed control scheme are explained. The validity of the proposed system is verified by simulation and experimental results.
2. Phase Current Analysis of Traditional Four Switching Patterns
The FSTPI topology consists of 4 power switches that provide two of the inverter output phases. The third phase is fed by the dc link from the center of a split-capacitor, as shown in Fig. 1.
Fig. 1.Power Circuit of FSTPI
In FSTPI system, there are four possible switching patterns to generate three-phase currents, as shown in Fig. 2 with ideal switches; these four switching patterns are V(0101), V(1010), V(1001), and V(0110), where “0” means the switch is turned off and “1” the switch is turned on in inverter. In the same figure, the free- wheeling diodes are ignored. In six-switch inverter, two zero voltage space vectors cannot supply the dc-link to the load, therefore no current flows through the load. The main difference of the four-switch inverter compared to its counterpart six-switch one is that one phase is always connected to the center tap of the split capacitors, so that there will always be current flowing through that phase even with voltage vectors V(0101) and V(1010), as shown in Fig. 2. Under balanced load condition with four-switch topology, there will be no current flow through the phase which is connected to the midpoint of the split capacitors using two possible non-zero voltage vectors, V(1001) and V(0110), as shown in Fig. 2. When voltage vectors V(1001) and V(0110) are used and the load is not completely balanced, only the resultant current of the other two phases flow through the phase connected to the midpoint of the split capacitors. This phenomenon will cause the ripple of current and torque. The simulation curves of current under 15rpm and 5Nm load using four-switching pattern is shown in Fig. 3.
Fig. 2.Conventional four-switch voltage vector topology
Fig. 3.Current curves using four-switching pattern
In addition, three different cases may be found at the commutation moment:
Case A: decaying current (id) vanishes at the same time rising current (ir) reaches to its final value +I in Fig. 4(a).
Case B: decaying current (id) vanishes before rising current (ir) reaches to it final value +I in Fig. 4(b).
Case C: rising current (ir) reaches the value +I before decaying current (id) vanishes in Fig. 4(c).
Fig. 4.Current behaviors during commutation
In case B and case C, current disturbance flowing through the phase connected to the midpoint of the split capacitors will cause distortion of magnetic linkage and pulsation of torque.
To solve above problems, this paper proposed a novel space vector modulation method and an optimal switching table.
3. DTC of BLDCM Using FSTPI
The typical mathematical model of a three-phase BLDCM is described as:
Where, νao , νbo , and νco are phase voltages; Rs is stator resistance; Ls is stator inductance ; ia, ib, and ic are phase currents; eao, ebo, and eco are phase back-EMFs.
3.1. Novel space vector modulation method
Generating a 120 electrical degree current conduction is inherently difficult with the conventional four-switch topology, because a BLDCM with non-sinusoidal back- EMF (i.e. trapezoidal) requires a quasi-square wave current profile to generate constant output torque compared to that of a permanent magnet synchronous motor with sinusoidal back-EMF requiring sinewave current. These currents which have 120 electrical degrees conduction period are synchronized with the flat portion of the corresponding phase back-EMFs, therefore a smooth electromagnetic torque can be obtained. As a result, at every instant of time only two phases conduct and the other phase is supposed to be inactive.
Although four voltage vectors in conventional fourswitch inverter system are sufficient enough to control the three-phase ac motors using PWM techniques, it will cause the ripple of current in BLDCM drive, which is shown in Fig. 3. So, additional voltage vectors are required for BLDCM with two phase conduction mode in order to control the midpoint current of the split capacitors at a desired value [8, 9]. Since the conventional method cannot provide a two- phase conduction method completely, a new modulation scheme with new switching patterns should be developed such that only two of the three motor phases conduct.
To obtain the switch modes of operation in four-switch BLDCM drive, a novel voltage vector selection look-up table is designed as shown in Table 1 and the Table 1 is concluded by the four-switch voltage vector topology which is shown in Fig. 6. Based on the Table 1, implementation of the voltage space vectors is depicted in Fig. 5
Table 1.The controlling voltage space vector and switch combination
Fig. 5.Voltage space vector chart (Vx=Vao+Vbo+Vco, x= 0,1,2,3,4,5,6,7)
3.2. Optimal switching table
Traditional DCC (direct current control) technique employs current hysteresis control [3, 10]. Normally, sixpossible voltage space vectors of four-switch topology are supposed to be used in Table 2 as shown in Figs. 6(a)-(f) similar to the six-switch version, however two of the voltage vectors V2 and V5 as shown in Fig. 6 create problems in the current control. When they are directly used in the voltage vector selection table (Table 2), back- EMF of the uncontrolled phase (phase-a) generates undesired current and distortions occur in each phase current. As a result, undesired electromagnetic torque is inevitable. Therefore, when the rotor position is in the sec.1, 2, 4, 5, 7, 8, 10 and 11, special switching pattern should be adapted
Fig. 6.Four-switch voltage vector topology
Table 2.Four-Switch voltage space vector selection for BLDCM drive
Additional two voltage vectors V0 and V7 which are unused in conventional four-switch PWM scheme are included in the optimal switching table to obtain torque in two-phase conduction four-switch BLDCM drive. The reason is there will be always current trying to flow in phase –a due to its back-EMF and the absence of switches controlling its current. As a result, there will be a distorted current in phase–a as well as in phase–b and –c. Therefore, voltage space vectors of phase–b and –c conduction can be difficult to implement for BLDCM drive unless some modifications are applied to overcome the back-EMF effect of the phase – a in these conditions. Selecting the right switching pattern to control the current on phase–b and – c independently will reduce the distorted currents on those phases and result in a smoother overall current and electromagnetic torque.
In order to improve the above problems, a DTC method has been proposed as shown in Fig. 7 and explained in detail below :
Fig. 7.Schematic of BLDCM control system
The DTC method controls instantaneous torque and flux to achieve high-performance operation [11-13]. For this purpose, an optimized switching table must be defined based on the output states of the instantaneous stator flux magnitude hysteresis controller and the electromagnetic torque hysteresis controller, together with the equivalent sector, in which instantaneous stator flux space vector is located [14, 15]. The outputs of hysteresis controller for torque and flux are shown in (3) and (4).
where Δλs is the stator flux hysteresis bandwidth and ΔTe is the torque hysteresis bandwidth. Since there are eight available switching states in FSTPI, stator flux (αβ) plane is divided into twelve different sectors spaced by 30 electrical degrees and an optimized lookup table (see Table 3) is used to translate these two control states, together with the stator flux position information defined in equivalent sector, to the inverter gate drive signals [16, 17].
Table 3.Voltage vector selection in sectors II and V for four-switch BLDCM drive (CCW)
In the following, how to eliminate flux disturbance and torque pulsation is explained, when rotor position is in sec. 2 and in CCW direction.
Case A(τT = τλ = 1 ): in this case, and , it should increase flux and torque, and V5 is selected.
Case B(τT = 1 and τλ = 0): in this case, and , it should increase torque and decrease flux, and V3 is selected.
Case C(τT = 0 and τλ = 1): this case is in contrast to the case B, and it should select V6.
Case D(τT = 0 and τλ = 0): this case is in contrast to the case A, and it should select V2.
In the same way, it can be analyzed, when rotor position in other sectors.
In addition, the direction of the rotor is important to define the specific switching pattern. If the rotor direction is CW, then the above claims are reversed.
This technique has the advantage of not requiring an integration step in the estimation calculation, thus removing problems associated with drift and integral windup [18, 19]. However, it does not rely on the motor parameters and an expensive position sensor.
In the next section, stator flux observation using single neuron adaptive PID is proposed and the observation principle will be analyzed in detail.
4. Flux Calculation
In the αβ reference frame, the model of flux can be described as:
where λs = [λsα λsβ ]T is the stator flux matrix.
The flux errors are given by:
Where eλ =[ eλα eλβ]T is the stator flux error matrix, denotes the estimated quantities, νs =[νsα νsβ ]T is the stator voltage matrix, is = [isα isβ ]T the stator current matrix.
Based on (6), flux observer is build using single neuron adaptive PID in Fig. 8.
Fig. 8.Schematic of observer of λs
In Fig. 8, the meaning of each part is explained as follows:
After discretization of (6), the input of single neuron adaptive PID is
Assuming the Quadratic performance index function is
In order to make modification of weighted coefficient wi=1,2,3 decrease along the reduce direction of E(t) , and (9) can be concluded.
Learning algorithm is as follows
Where, K is neuron proportional coefficient, wi=1,2,3 is the weighting efficient of xi=1,2,3, 0 < wi=1,2,3 < Umax , and Umax is maximum amplitude.
Single neuron weighting efficient can be expressed
where, ηp, ηI, and ηD are proportional, integral and differential weights of learning rate, respectively.
PID parameters are
5. Stability Analysis and Parameter Selection
5.1 Lyapunov stability analysis
Define a Lyapunov candidate function
In the learning process, the change of ν(t) can be expressed as follows
Assuming e(0) = 0 , (15) can be concluded.
The change of ν(t) can also be expressed as follows
H = [∂e(k) / ∂w(k)]T = (∂e(k) / ∂w(k))[∂u(k) / ∂w(k)]T , (17) can be concluded.
Substituting (17) in (15) yields
For global asymptotic stability, Δν(k) < 0 . Hence, the following equations can be deduced:
So, the scope of the learning rate is
For Δν(k) < 0 , (21) can be concluded from (14)
So learning algorithm is convergence, and control system is stability.
5.2 Parameter selection
When the error is large, proportional coefficient K should large for having fast start-up speed. And when the error is small, K should small for preventing overshoot. Generally, nonlinear adjustment formula for K is shown in (20) .
Where k0 is steady-state proportional coefficient, A is regulation coefficient.
But a critical problem of the low-speed operation of the flux observer is the dc-measurement offset. In reality, να and νβ are reconstructed from the dc-bus voltage Vdc and the switching status. Any dc-measurement error, switch voltage, and dead-time voltage are reflected in both να and νβ , especially in low-speed performance. In Fig. 9, torque and flux estimation error curves are provided under given speed 1500rpm and 3Nm load conditions. In Fig. 9(a), it is obvious that the operational performance of control system is poor.
The effects of the dc offset can be alleviated by introducing a variable PI component in regulation coefficient as follows:
Where kp = Rs / Ls , and ki = 1/ Ls . A ≈ k0 /10 is practical for a real drive.
The advantages of (23) is the function of last item can be obviously increased, when the error is large at the beginning operation or flux error is large, especially at lowspeed. With the increasing of speed or flux error becoming small, the function of last item is obviously weakened, and the overshoot is limited. After compensation, the result is shown in Fig. 9(b).
Fig. 9.Torque and flux estimation error
6. Simulation and Experiment Results
According to above analysis, it can get the expressions of rotor position, speed and electromagnetic torque as follows.
Direct torque control system structure of BLDCM is shown in Fig. 7. Based on Fig. 7, simulation model has been built using Matlab/Simulink, and parameters of BLDCM are shown in Table 4. Simulation results are shown in Fig. 10.
Table 4.Parameter of BLDCM
Fig. 10 describes the simulation results which is control system starts up with 5Nm load, and load sudden drops from 5Nm to 1Nm. From Fig. 10(a), it can be seen flux observation curve is smooth, and good observation effect of flux can provide accurate information for system control. In Fig. 10(b) and (e), current curve of DTC is smoother than that of DCC, and DTC can improve the current distortion causing by uncontrolled phase. In Fig. 10(c), (d), (f), and (g), it is shown that the DTC can effectively alleviate commutation torque ripple, and improve the torque and speed response.
Fig. 10.Simulation curve
Based on the above research content, BLDCM control system has been established with IPM (Intelligent Power Module) devices, and dsPIC6010A is used as main controller. Experimental results are shown in Fig. 11 and 12 under torque changed from 5Nm to 1Nm at high speed (1500rpm) and low speed (15rpm). Fig. 13 shows the experimental results under traditional DCC, when torque is changed from 5Nm to 1Nm at high speed (1500rpm) and low speed (15rpm).
In Fig. 11(a) and Fig. 12(a), since the dc voltage compensation has been introduced, flux observation precision has been improved and it can provide accurate flux size and position information for the DTC.
From Fig. 11(b), Fig. 12(b) and Fig. 13(a, b), it can be seen that the DTC can improve current distortion causing by uncontrolled phase, and the current response speed is faster than that of traditional DCC.
From Fig. 11(c, d), Fig. 12(c, d) and Fig. 13(c, d, e, f), it can be seen that the DTC can improve torque pulsation, and the DTC has better dynamic and static performance characteristics of torque and speed than that of traditional DCC.
Fig. 11.Experimental curves for DTC under high speed
Fig. 12.Experimental curves for DTC under low speed
Fig. 13.Experimental curves for traditional DCC
In this paper, a novel voltage space vector modulation scheme, an optimal switching table and a single neuron stator flux observer for direct torque control were proposed. The flux observer does not require speed adaptation and is not susceptible to speed estimation errors, especially, at low speed. The stability of the proposed observer is proven by the Lyapunov stability analysis. The effects of dc-measurement offsets are mitigated by incorporating an integral compensating term in the observer gain. The proposed single neuron flux observer is capable of delivering high performance over a wide speed range, including very low speeds. As a consequence, current distortion and torque pulsation are effectively improved, and the response speed of speed and torque are significantly improved. The proposed DTC can also be used in servo control for BLDCM.