Sensor fusion: The application of softcomputing in monitoring and control for railroad maintenance

Although the potential of sensor data fusion as a way to improve situation awareness nowadays is widely regarded as a valuable tool by the monitoring and control community, there is still little quantitative evidence available that supports the acclaimed benefits of generic fusion methodologies and thus justifies their application. It is therefore legitimate to ask how much value sensor fusion adds to the observation process. In this paper we will try to identify in what situation the application of sensor fusion techniques may be beneficial and which sensor fusion techniques seem to be among the most promising. The starting point of our discussion is the observation that the added value of fusion of similar sensors must originate from a nonlinear combination of sensor data streams. This observation naturally gives rise to the application of novel nonlinear models, e.g. from the area of soft computing, viz. fuzzy logic, neural networks and evolutionary programming. At Strukton a sensor fusion approach for railroad maintenance management has been adopted, in which an integrated approach towards sensor optimization, sensor management, and early fusion is pursued. Only in this way one may hope to attain the goal of sensor fusion, viz. an improved situation assessment and early warning to trigger preventative maintenance. Concentrating on the primary information sources, i.e. the sensors, and using their synergy at the earliest stages of signal processing, inevitably forces one to invest in state-of-the-art data-acquisition and real-time (pre)processing facilities.