Artificial Multi-Layer Neural Network Including Feedback Connection based on Causality

M. Ohka, K. Esumi, and Y. Sawamoto (Japan)


Neural networks, Multi layer, Feedback connection, Causality, Piezoelectric actuator


Since the relationship between applied voltage and displacement, which is caused by the voltage, shows a hysteresis loop in the bimorph piezoelectric element, we produced a control system for a piezoelectric actuator based on a multi-layer artificial neural network to compensate the hysteresis. A neural network, which is formulated based on causality, is comprised of four neurons in the input layer, ten neurons in the hidden layer, and one neuron in the output layer. The output neuron emits the voltage time derivative, which is determined by the current values of displacement, voltage, and the increment or decrement state. A two bits signal expressing the increment or decrement condition is generated by two input neurons. Other two input neurons calculate the current voltage and displacement values. In the learning process, the network learns the hysteresis including the minor loops. After learning the hysteresis loops including the minor loops, the neural network simulates these hysteresis phenomena with very high accuracy. A series of verification tests was performed on a parallel-type two-dimensional actuator composed of two bimorph piezoelectric elements and two small links connected by three joints. The network provided a voltage history for several sizes of circular trajectories in two dimensional space.

Important Links:

Go Back

Rotating Call For Paper Image