Detection of discrete events in real time vibration data using sensor networks for intelligent environments

Events or changes detected in real time data can be used for many tasks in intelligent environments from monitoring and tracking to localisation and grouping. In this paper we look at methods for detecting these events in real time vibration data obtained from mechanical machines. Each machine has its own vibration cycles dependant on application and their unique traces can be reduced to a series of discrete events, these can then be used for a variety of uses including our focus of logically grouping the sensor nodes. We argue that as vibration data is essentially low frequency sound we can apply methods used in speech recognition and audio processing to this domain. We then compare these audio based methods with standard change detection techniques and a method we have derived from an audio segmentation technique. Our experimentation shows that our method out performs the change detection and power signature methods and that it is more computationally suitable to embedded sensing. We move on to show that our algorithm can indeed be run on an computationally simple device, a SunSPOT from Sun Microsystems Inc., and that it can detect events from the vibration data stream in real time.