Acquisition of sensor fusion rule based on environmental condition in sensor fusion system

The manufacturing systems have become more and more complex for adapting to various process conditions. Recently, various and numerous sensors are equipped in the systems for measuring various states in processes. For efficient manufacturing, a sensor fusion method is needed for inferring state which cannot be measured by conventional sensors. So, many sensor fusion methods have been proposed so far. We propose a sensor fusion method with sensor selection based on the reliability of sensor value. However, conventional sensor fusion methods cannot infer states accurately under various environmental conditions. In this paper, we propose a sensor fusion system with a knowledge database for fusing under various environmental conditions. The sensor fusion rules under each environmental condition are stored in the knowledge database. Then, the system selects sensors according to an appropriate sensor fusion rule in the knowledge database and fuses selected sensor values by a recurrent neural network. Additionally, the system generates a new sensor fusion rule for an unknown environmental condition by the genetic algorithm. For showing the effectiveness, we apply the proposed method to inference of the surface roughness in the grinding process.