Seven-level NPC Inverter-based Neuronal Direct Torque Control of the PMSM Drives with Regulation Speed Using Neural PI Controller

In the scope of this work, a new structure of the direct torque control (DTC) technique scheme of permanent magnet synchronous motors (PMSMs) is proposed using a seven-level neutral point clamped (NPC) inverter and artificial neural networks (ANNs) controllers. This intelligent strategy was used to replace, on the one hand the integral proportional (PI) speed controller, on the other hand and the traditional switching table in order to reduce total harmonic distortion (THD) of voltage/current, stator flux ripple and electromagnetic torque ripples. The modeling and simulation of the proposed control scheme were carried out in Matlab/Simulink environment. The simulation results show that the proposed strategy scheme reduced harmonic distortion of stator voltage/current, stator flux and electromagnetic torque ripples compared to traditional DTC control scheme, DTC control scheme with space vector pulse width modulation (SVPWM) and three-level DTC control scheme. It was found that the stator current waveform becomes purely sinusoidal with a reduction in the harmonic distortion rate to 11.77%.


Introduction
PMSMs for Permanent Magnet Synchronous Motors are one of the most proper and efficient motors in electricity industry. these motors are good candidate for applications such as naval and space systems, electric vehicles and etc. The most important advantages of such motors are: high efficiency and power density, low loss, low maintenance cost, easy construction and etc [1]. Various control strategies have been proposed for regulate the electromagnetic torque of the PMSM drives. Most existing techniques used a field oriented control (FOC) [2], robust control [3], intelligent control [4], and DTC control [5]. Stator FOC control using proportionalintegral (PI) regulators is the traditional strategy used for PMSM drives. However, PI regulators performance is highly dependence on tuning of parameters and accurate tracking of angular information of stator voltage/flux. Moreover, the FOC control strategies need accurate values of machine parameters and rotor speed.
The DTC strategy has been introduced in the 1980s by Noguchi and Takahashi as an alternative to FOC control, with the two ofold objective of simplifying the control algorithms and achieving similar or even better performances [6]. In this strategy, two hysteresis comparators, namely flux and torque controllers are selected to determine the inverter instantaneous switching state. The DTC strategy is simple and alternative approach technique formulation that does not require decomposition into symmetrical components.
To ensure the robustness and good performance of the DTC strategy using hysteresis comparators, several approaches have been recently proposed. In [7], the authors proposed a new DTC strategy based on sliding mode controller (SMC) to control the PMSM. In [8], the authors proposed a DTC strategy with space vector modulation (SVM) to regulate the electromagnetic torque and stator flux ripples. DTC and fuzzy logic controlled (FLC) are combined to control PMSM [9]. SMC and fuzzy logic are combined to control PMSM [10]. In [11], a seven-level neural DTC with Regulation Speed Using Neural PI Controller fuzzy speed controller of induction motor (IM) was presented. A four-level DTC control with ANN is proposed to regulate the stator flux and electromagnetic torque of the IM drives [12]. DTC and NN controller are combined to regulate the torque of PMSM [13]. In [14], the modified DTC strategy which is based on fractional sliding mode has been proposed. In this paper, we propose a new DTC control scheme based on ANN method of a PMSM fed by seven-level inverter. The simulation studies have performed with Matlab/Simulink board to performance testing of the proposed DTC algorithm. The section II is dedicated to the modeling of the PMSM machine. In section III of this paper, the basic principles of the conventional DTC control scheme has shortly introduced.
In section IV, detailed information of the proposed strategy based on ANN controller has presented. Section V presents the simulation results of the both strategies. Finally, conclusion has given in section VI.

Model of PMSM
The mathematical model of the three-phase PMSM can be expressed in the (q-d) synchronous rotating frame by the following nonlinear equations [15,16]: Where: p: is the number of pole paire. Te: is the electromagnetic torque So the dynamic equation is given as: Where: Ω: is the mechanical rotor speed. J: is the moment of inertia. Tr: is the motor torque load. f: is the coefficient of viscous friction.

Traditional DTC Control
The goal of the DTC control is to control the electromagnetic torque and stator flux by the application of the optimal voltage vector synthesized from the determination of the suitable sequences to be applied to the power switches devices of the two-level inverter [17]. The principle of the DTC strategy is illustrated in Figure 1. The estimated value of electromagnetic torque and stator flux are compared with references values using the hysteresis comparators [18]. The stator flux can be obtained, in a stationary reference frame (α, β), as follows [19,20]: The stator flux amplitude is given by: The stator flux angle is calculated by: The stator flux angle is calculated by: The stator voltage is calculated by: And: The stator current is calculated by: With: Is = Isα + j Isβ Electromagnetic torque (Te) can be calculated using the components of the estimated flux and measured currents: The representation of the space voltage vectors of the traditional inverter for all switching states is given by Figure  2. In traditional DTC a two-level hysteresis comparator is used for the stator flux (HSF) and a three-level hysteresis comparator for the electromagnetic torque (HET) (See Figure  3 and Figure 4). Finally, based on the values of constants HSF and HET and the position of the stator flux (six region control), the inverter switching algorithm is as shown in Table 1 [21].

ANN Control Based Speed Controller
The block diagram for neural based proportional-integral (PI) regulator is shown in Figure 9. The neural PI (PIN) controller design is based on intuition and simulation. The training used is that of the retropropagation of Levenberg-Marquardt (LM). The Table 5 shows the parameters of ANN controller.  Such a neural controller contains three layers: input layers, hidden layers and output layers. Each layer is composed of several neurons. The block diagram of the ANN controller of the speed regulation is given by Figure 10. The structure of layer 1 and layer 2 is shown in Figure 11 and Figure 12 respectively.

ANN Based Switching Table
In the conventional DTC strategy, stator flux and electromagnetic torque can be directly controlled by using hysteresis comparators and switching table. In this approach, we replaced the switching table by a single ANN controller with stator flux hysteresis comparator, torque hysteresis comparator and sector as inputs. The switching (Sa, Sb and Sc) of the inverter will be the control output. The proposed ANN controller is the retropropagation of Levenberg-Marquardt (LM). The parameters of the LM algorithm is shown in Table 6. The structure of neural switching table is illustrated in the Figure 13.

Simulation Results
The simulation results of seven-level DTC-ANN with neural speed controller (DTC-ANN-PIN) of sensorless PMSM are compared with traditional DTC (DTC-2N) strategy. The performance analysis is done with harmonic distortion of stator current, sector, torque, flux and speed. The stator flux and torque references used in the simulation results of the classical DTC and DTC-ANN with the neural speed controller are 0.3 wb and 5 N.m→-5 N.m respectively. The machine is running at 100 rad/s→-90 rad/s→100 rad/s. The PMSM used in this case study is a 1,5Kw, 220/380V, three poles, 50Hz, with the following parameters: Rs= 1.4Ω, Lq=0.0066 H, Ld=0.0058 H, Ѱf=0.1546 Wb.
The system has the following mechanical parameters: J = 0.00176 Kg.m 2 , fr= 0.00038818 Nm/s.  Figures 20-21 shows the THD of stator current of the PMSM obtained using FFT (Fast Fourier Transform) method for conventional DTC strategy and seven-level DTC-ANN with neural speed controller one respectively. It can be clearly observed that the THD is minimized for seven-level DTC-ANN-PIN when compared to conventional DTC control scheme. Table 7 shows the comparative analysis of THD value.
For the conventional DTC strategy and seven-level DTC-ANN with PIN controller, the speed, stator flux and electromagnetic torque, tracks almost perfectly their references values (see .
Electromagnetic torque response comparing curves are shown in Figure 17. See figure the torque ripple is significantly reduced when the seven-level DTC-ANN with neural speed controller is in use. Figure 18 shows the flux responses of both the conventional DTC strategy and seven-level DTC-ANN with PIN controller. It is found that the proposed DTC exhibits smooth response and lesser ripple in flux as compared to the conventional DTC strategy and three-level DTC control scheme in ref. [22].

Conclusion
In this paper, the seven-level neural DTC with neural PI controller is presented and it is compared with traditional DTC control. The simulation results obtained for the proposed control scheme illustrate a considerable reduction in electromagnetic torque ripple, harmonic distortion of stator current and stator flux ripple compared to the conventional DTC utilizing two-level inverter and even compared with results presented in other recent works such as [14] and [22].