This paper presents and evaluates the concept and implementation of two distinct multi-sensor systems for the automated manufacturing based on parallel hardware. In the most sophisticated implementation, 12 processors had been integrated in a parallel multi-sensor system. Some specialized nodes implement an Artificial Neural Network, used to improve photogrammetry-based computer vision, and Fuzzy Logic supervision of the sensor fusion. Trough the implementation of distributed and intelligent processing units, it was shown that parallel architectures can provide significant advantages compared to conventional bus-based systems. The paper concludes with the comparison of the main aspects of the transputer and the DSP-based implementation of sensor guided robots.
Welding process; multi-sensor system; parallel processing; automation; artificial neural network
