904 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 37, NO. 3, MAY/JUNE 2001 Optimized Torque Control of Switched Reluctance Motor at All Operational Regimes Using Neural Network Khwaja M. Rahman, Member, IEEE, Suresh Gopalakrishnan, Member, IEEE, Babak Fahimi, Member, IEEE, Anandan Velayutham Rajarathnam, Member, IEEE, and M. Ehsani, Fellow, IEEE Abstract—Switched reluctance motor (SRM) optimal control parameters, which maximize torque per ampere, are calculated using a dynamic SRM model. In order to include the effect of the magnetic nonlinearity, static torque and flux-linkage data are used in the dynamic model. The static data are generated experimentally. To recreate these control parameters, online, artificial neural networks are used. Two separate networks are trained. One is trained with the low-speed control parameters for torque control at low speed, while the other is trained with the high-speed control parameters for torque control at high speed. The speed at which the SRM makes a transition from chopping control to single-pulse operation (i.e., low-speed to high-speed op- eration), commonly referred to as base speed, is torque (current) dependent. A small table is maintained in the controller to identify the base speed for different torque demands. When the motor exceeds the base speed for a certain torque demand, the controller switches from the low-speed neural network to the high-speed neural network and vice versa. It is also shown that the SRM is capable of producing an extended constant-horsepower operation with this optimal control. The power factor (the energy ratio) is shown to improve in this extended speed constant-horsepower range. Simulation and experimental results are presented to demonstrate the effectiveness of the proposed control scheme. Index Terms—Electrical drives, neural network control, switched reluctance machine. Paper IPCSD 00–065, presented at the 1998 Industry Applications Society Annual Meeting, St. Louis, MO, October 12–16, and approved for publication in the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Industrial Drives Committee of the IEEE Industry Applications Society. Manuscript submitted for review June 1, 1998 and released for publication January 22, 2001. K. M. Rahman was with the Power Electronics Laboratory, Department of Electrical Engineering, Texas A&M University, College Station, TX 77843-3128 USA. He is now with General Motors Advanced Technology Vehicles, Torrance, CA 90505 USA (e-mail: rahmank@pcssmtp.hac.com) S. Gopalakrishnan was with the Power Electronics Laboratory, Depart- ment of Electrical Engineering, Texas A&M University, College Station, TX 77843-3128 USA. He is now with the Mechatronics Department, Delphi Automotive Systems, Shelby Township, MI 48316 USA (e-mail: suresh.gopalakrishnan@delphiauto.com). B. Fahimi was with the Power Electronics Laboratory, Department of Elec- trical Engineering, Texas A&M University, College Station, TX 77843-3128 USA. He is now with Electro Standards Laboratories, Cranston, RI 02921 USA (e-mail: bfahimi@ElectroStandards.com). A. V. Rajarathnam was with the Power Electronics Laboratory, Department of Electrical Engineering, Texas A&M University, College Station, TX 77843-3128 USA. He is now with the Advanced Motor Electronics Group, Emerson Electric Company, St. Louis, MO 63136 USA. M. Ehsani is with the Power Electronics Laboratory, Department of Electrical Engineering, Texas A&M University, College Station, TX 77843-3128 USA (e-mail: ehsani@ee.tamu.edu). Publisher Item Identifier S 0093-9994(01)03936-6. I. INTRODUCTION T HE switched reluctance motor (SRM) is gaining much at- tention due to its simplicity, low cost, ability of very-high- speed operation, and safe and hazard-free operation. However, the nonlinearity in the operation of the SRM complicates the analysis as well as the control of this motor. The nonlinearity arises due to its, by design, saturation region of operation in al- most all of its entire operational regimes. Several attempts have been made to model the nonlinear magnetic characteristics of the SRM. Some are based on analytical formulations [1]–[4], the others are based on finite-element method [5]–[8]. The former suffer from lack of accuracy, while the latter, although to some extent accurate, are complicated and very time consuming. Both of these methods are frequently used for design optimization of the SRM, however, they are too detailed to be implemented in real time for control purposes. The capability to accommodate nonlinear modeling has made artificial neural networks (ANNs) ideal candidates to solve the control strategies of an inherently nonlinear system [9]. A self-organized Kohonen neural network has been presented in [10] for the modeling of nonlinear SRM torque characteristics as a function of position and current. Nonlinear modeling of the SRM based on the backpropagation neural network is presented in [11]. Recently, torque-ripple minimized control of an SRM using ANNs has also been presented [12]. In all these approaches, the neural nets were trained using static magnetization data. Due to the static nature of the solution, these methods perform inadequately in the dynamic regime of the SRM operation. Moreover, the solutions provided by these methods are not optimal. An optimal torque control of SRM using a neural network is presented in [13]. This control approach attempts to minimize torque ripple with maximized torque per ampere. This is achieved in this control approach by profiling the phase current. However, only low-speed operation was considered in this work. A self-tuning control of an SRM is presented in [14] which tries to maximize torque per ampere. However, this method can only be implemented at steady state. A neural-network-based self-tuning control of an SRM is also presented in [15]. This method, which was implemented for low-speed operation of the SRM, controls the phase turn-off angle online to maximize torque per ampere. This method was extended later in [16] to accommodate the high-peed operation 0093–9994/01$10.00 © 2001 IEEE