Real-time monitoring and diagnosing of robotic assembly with self-organizing neural maps

An application of a self-organizing neural-network map is presented for real-time execution monitoring and diagnosing of robotic assembly. The self-organizing map has the ability to spontaneously react to changes in dynamic assembly processes. It offers simple and flexible ways of organizing diverse assembly interactions between tools, parts, robot, and sensory data. A number of different types of multi-dimensional maps are described for various combinations of assembly interactions. Limitations of the approach and possible solutions are discussed. The performance of the approach is demonstrated on a sample assembly. Some observations and insights gained during the neural-network training phase are included