| |
MICROCONTROLLER OPTIMAL IMPLEMENTATION OF A NEURAL NETWORK POSITION ESTIMATOR FOR A VARIABLE RELUCTANCE LINEAR ACTUATOR
| [View]
[Download]
|
| Author(s) |
S. Cincotti; M. Marchesi; A. Serri |
| Abstract |
A sensorless control for a variable reluctance linear actuator implemented on a lowcost microcontroller (MCU) is presented. The motor position estimator is implemented by an artificial neural network (ANN) properly trained. The ANN estimates the motor position by
measuring the ripple in the motor phase during chopping-mode supply. The feasibility of the position estimator is preliminary assessed by using a simplified mathematical model of the
motor. Then, the neural design is adapted to fit experimental measures of the motor behavior. Finally, the ANN design is optimized for the 8-bit fixed point arithmetic of the MCU. To obtain
faster implementation of the ANN, discrete optimization of weight values is constrained to sum-of-power-of-two in order to convert multiplication into shift and add operations. Sigmoid
outputs are obtained by a fast look-up table. Numerical and experimental results are presented to evaluate the sensorless drive behavior. |
| Download |
| Filename: | Unnamed file |
| Filesize: | 612.7 KB |
|
| Type |
Members Only |
| Date |
Last modified 2016-01-08 by System |
| |
 |