Published Paper Details:

This paper proposes an alternative design of magnetic encoder for analog angle measurement based on Radial Basis Function (RBF) neural network. In order to realize analog angle output, multiple linear Hall-effect sensors and a cylindrical magnet are used. RBF neural network is used to approximate the multi-dimensional nonlinear function between the vector of sensor values and the angle output. For successful approximation of the function between inputs and output, the parameters of the RBF network needs to be determined by training. Training data consisting of vector of sensor values and the corresponding angle value at various angular positions of the rotatory shaft needs to provided as input to the RBF neural network trainer. The RBF neural network with trained parameters can be used to output the angle value for the given sensor values as input. The trained RBF neural network operation is implemented using 8-bit microcontroller. With this design and implementation, the angle value at various positions of the rotatory shaft is obtained with sufficient level of accuracy. This design of the magnetic encoder also allows some flexibility in terms of placement of the sensors and magnet.