This paper describes the performance of a novel tactile sensing system able to determine a description of a load in contact with a surface. The approach uses sensing elements in order to aim for mechanical simplicity and to formulate the load description. It integrates a deflecting surface, sensing elements, and computer interpretation algorithm. The description of the contacting load is derived in terms of a continuous function outputting position, width and weight. To demonstrate the principles clearly, the experimental surface selected for this investigation is a one-dimensional (1-D) simply supported beam that uses eight proximity sensors to detect deflection in response to the applied load. Back propagation neural networks are used in a cascaded architecture to process the sensory data to automatically estimate load position and width both within an error of 4% and 8% of the full range, respectively. The total value of the load is determined to within an error of 4% of the true value.
|Number of pages
|IEEE Transactions on Instrumentation and Measurement
|Published - 7 Aug 2002
- Measurement error
- Neural network application