Digital control of a servo system using neural networks

Many types of industrial drives require high performance controllers to ensure good dynamic performance in increasingly complex environments. One promising method to implement such a controller is to use neural networks which possess inherent self-learning, nonlinear mapping and fault-tolerant capability. However, a major issue with the application of neural network controllers is the question of how to optimally train for an effective control performance. Unfortunately since there is no systematic and universally effective method available for training, sometimes the performance of a “trained” neural network controller turns out to be significantly worse than expected, and much work remains to be done in this area. This paper presents the application of a newly established technique for training neural networks based on plant response, to develop a high performance brushless DC motor servo system where the position control is implemented by a neural network. The system uses a master/slave DSP structure to achieve the fast calculations required to realize the complex control algorithms, and interfaces to the outside world through a low cost 8 bit microprocessor which acts as an I/O front end master. The experimental results confirm the superior performance of neural network controllers and the effectiveness of the proposed master-slave architecture to implement a digital version of such a controller