Sensing positions optimisation of a distributive tactile sensor using principal component analysis

This paper describes a method to optimise sensing positions of a two-dimensional distributive tactile sensor for determining an applied load position from surface deflections. The distributive approach relies on coupling between sensing positions that capture a pattern of response to contacting load. The paper describes an experimental arrangement and the corresponding mathematical model that incorporates surfacepsilas response induced by a contact. The size of the experimental rig is 250 mm times 340 mm. The determination of an applied load position is completed through a back propagation neural network as an interpretation algorithm using surface deflections as input data. The average Euclidean error using 16 inputs from measurement was approximately 23 mm when sensing positions were at an equal pitch. Optimisation was achieved using principal component analysis as a tool to evaluate the performance. The number of inputs was simulated surface deflection at 4-16 positions. It was found that the number of sensing elements converged accordingly to the number of principal components (eigenvalues) used in optimisation. In terms of performance, the errors ranged from approximately 23.9-15.8 mm and 20.5-14.0 mm when inputs were mathematically derived from 4-16 non-optimised and optimised sensing positions respectively. Optimisation was an effective method to enhance the accuracy in determining an applied load position, in particular with a smaller number of sensing elements.