Sensing health: fault detection, isolation, and minimization of on-body sensors

Abstract

Monitoring activities of daily living of senior citizens especially those who live alone as well as those who have certain motion disorders such as Alzheimer's and Parkinson's is increasingly possible and very useful for healthcare purposes. This kind of pervasive and remote monitoring is possible due to availability of "on-body" sensors that combine sophisticated inertial sensing, wireless communication and signal processing technologies. The inertial sensors being electro-mechanical in nature are susceptible to errors and faults. In typical wireless sensor systems, such errors and faults are overcome by deploying large number of sensor nodes and taking a majority of the sensed values. However, in the case of on-body sensors, it may be highly inconvenient for an elderly person to wear a large number of sensors. Similar to on-body sensors, tiny electro-magnetic sensors are beginning to be used for tracking tongue movements during speaking. This breakthrough technology can give insights into how we as humans produce sound and these insights can help better train speech impairments. Considering the inconvenience in using such tongue sensors, it will be important to minimize the number of tongue sensors needed for a particular requirement. In this talk, we address the issue of identifying and isolating faulty sensors as well as minimize the number of needed sensors for remote monitoring of daily living activities and tongue movement tracking. We discuss issues such as: (i) faulty node identification when a small number of body sensors is deployed; (ii) identification of such a faulty node based on sensor node groups that share similar sensor signal patterns; and (iii) minimize the number of sensors needed for a particular application.