Health condition monitoring often entails monitoring and detecting changes in the structure of associated random processes. A common trigger for an alarm is when the process amplitude exceeds a specified threshold for a certain period of time. A less common trigger for an alarm is when the process bandwidth changes significantly. This latter type of change occurs, for example, in EEG just prior to the onset of an epileptic seizure [Sherman (2008)]. One can monitor the process directly, or one can convert the process to a 0/1 process where 0 denotes ‘within’ and 1 denotes ‘outside of’ a specified tolerance. Such a process is given many names. One is a binary process, another is a Bernoulli process. If the underlying process is a Gauss-Markov process, then the associated 0/1 process becomes a Markov 0/1 process. The main parameters associated with such a process are the following probabilities: (i), (ii), and (iii). In this work we use the variance expressions for the estimators of these probabilities that were reported in Sherman (2011), in order to detect changes with specified false alarm probabilities. We demonstrate their value in detecting bandwidth change via zero-crossing estimates, and detecting amplitude change via threshold excursion estimates.

This content is only available via PDF.
You do not currently have access to this content.