A comparison of artificial neural network and extended Kalman filter based sensorless speed estimation

• This study focused on the sensorless speed estimation of a dc-motor.
• The speed estimation algorithm does not require any sensor.
• The ANN results was compared with EKF results to show performances of algorithms.
• The NN estimates the speed of motor with low error and high accuracy than EKF.
• The algorithm is an alternative observer for motor speed control systems.

In industry speed estimation is one of the most important issue for monitoring and controlling systems. These kind of processes require costly measurement equipment. This issue can be eliminated by designing a sensorless system. In this paper we present a sensorless algorithm to estimate shaft speed of a dc motor for closed-loop control using an Artificial Neural Network (ANN). The method is based on the use of ANN to obtain a convenient correction for improving the calculated model speed. Three architectures of ANNs are developed and performance evaluations of the networks are performed by three performance criteria. After the evaluations, Levenberg–Marquardt backpropagation algorithm is chosen as learning algorithm due to its good performance. The speed estimation performance of developed ANN was compared with Extended Kalman Filter (EKF) under the same conditions. The results indicates that the proposed ANN shows better performance than the EKF. And ANN model can be used for speed estimation with reasonable accuracy.