AbstractThis paper presents an artificial neural network (ANN) trained on the patterns of slip signals; these patterns were generated by using conventional sensors with a novel design of fingertip mechanism for detecting the slippage of a grasped object under different types of dynamic loads. This design is to be used with an underactuated triple finger artificial hand based on the pulleys-tendon mechanism. The grasped object is designed in a prism shape with three direct current motors with unbalance rotating mass to generate excitation in the object. Also, this object is covered with different types of surface materials, namely, spongy rubber, glass, and wood. Three types of external loads are used to disturb the grasping process represented by quasi-static pulling on the object, the dynamic load on the object, and on the artificial arm in separate form. The mathematical modeling has been derived for the proposed design to generate the signal of contact force components ratio through using the conventional sensor signals with the aid of Matlab-Simulink software. The ANN has been trained on the basis of the patterns of force component ratio signals at slippage occurrence, in order to detect slippage and then prevent it without the need for any knowledge about the surface properties of the grasped object. The experimental results are discussed in comparison with the physical aspect of the slippage phenomenon, and they show good agreement with the physical definition of the slippage phenomenon. In addition, the network evaluation results are discussed with different parameters that govern the controller operation, such as network error, classification efficiency, and delay in response time.
