Ultrasonic motors are a good alternative to electromagnetic motors in medical robotics, since they are electromagnetically compatible. Estimating speed instead of using encoders reduces cost and dimension of the robot on the one hand and increases reliability on the other hand. However, no sensorless speed controller is yet industrialized. Analytical models of the traveling wave ultrasonic motor being too complex to be exploited for sensorless control purpose, we suggest speed estimation based on artificial neural networks. The artificial neural network is designed based on a sensitivity analysis using design of experiments methods. Factorial designs have been chosen to find out the effects of each input factor, but also the effect of their interactions. First results show that speed estimation using artificial neural networks is a promising approach. The artificial neural network optimized with design of experiments methods is a valid model of the traveling wave ultrasonic motor to estimate speed.