Applications of switched reluctance motor (SRM) to direct drive robot are increasingly popular because of its valuable advantages. However, the greatest potential defect is its torque ripple owing to the significant nonlinearities. In this paper, a fuzzy neural network (FNN) is applied to control the SRM torque at the goal of the torque-ripple minimization. The desired current provided by FNN model compensates the nonlinearities and uncertainties of SRM. On the basis of FNN-based current closed-loop system, the trajectory tracking controller is designed by using the dynamic model of the manipulator, where the torque control method cancels the nonlinearities and cross-coupling terms. A single link robot manipulator directly driven by a four-phase 8/6-pole SRM operates in a sinusoidal trajectory tracking rotation. The simulated results verify the proposed control method and a fast convergence that the robot manipulator follows the desired trajectory in a 0.9-s time interval.
TopIntroduction
There has recently been a considerable interest in developing efficient direct drive robot manipulator, because the elimination of gear boxes simplifies the construction of manipulators and removes sources of flexibility and nonlinearities, such as friction and backlash, which are known to cause difficulties in designing high-quality controls. Many motors, like conventional dc motor (Mohamed et al., 2004), brushless dc motor (Park et al., 2003), induction motor (Hsu et al., 2005), and switched reluctance motor (SRM) (Wallace et al., 1991; Spong et al., 1987; Chen et al., 2003; Amor et al., 1993; Bortoff et al., 1998; Milman et al., 1999), have been investigated in applications to direct drive manipulator, where the SRM is increasingly popular due to its simple structure, low cost and reliability in harsh environments. Moreover, the SRM can produce high torque at low speed, which is compatible with direct drive manipulator specifications.
SRM model, however, exhibits significant coupled nonlinear, multivariable, and uncertainty. Hence the classical linear control schemes cannot provide the required performances for high-precision position control.
The latest advances and engineering applications of cognitive informatics have gotten blooming achievements (Wang et al., 2010; Wang, 2009a, 2009b; Zhong, 2008). Artificial neural networks have been adopted extensively due to their abilities to achieve nonlinear mappings and fast autonomous learning, in a wide variety of domains, from modeling and simulation (Cai et al., 2011), rotor position estimation (Beno et al., 2011), classification of musical chords (Yaremchuk et al., 2008), sensorless control for a switched reluctance wind generator (Echenique et al., 2009), to robot control (Bu et al., 2009; Bugeja et al., 2009; Dierks et al., 2009, 2010; Ferreira et al., 2009; Hong et al., 2009; Hou et al., 2010; Tan et al., 2009; Wai et al., 2010; Wei et al., 2009; Zhao et al., 2009).