Increasing the Accuracy of Time-tabled Condition Monitoring

Many applications benefit from the ability to monitor the value of conditions over large, possibly distributed data sets. However, mechanisms to achieve this typically impose a performance overhead that the owners and users of such systems may be unwilling to accept. A compromise solution is to allow system owners to decide at what times they will allow computational resources to be given over to condition monitoring and at what times they will not. This allows the system owner to maintain maximum throughput rates during the most important periods of the day or week, while still reaping the benefits of a limited form of condition monitoring. The monitoring is “limited” because the results will contain inaccuracies. However, in some applications, even inaccurate information about conditions will be of value, if it can be provided at minimal cost to the organisation in question. This approach is called non-intrusive condition monitoring. In this paper, we discuss a variety of methods for reducing the inaccuracy of non-intrusive condition monitoring in distributed information systems. The methods employ a range of temporal reasoning techniques, and promise differing degrees of improvement in accuracy. They also each impose different costs on the operation of the system as a whole. We present some experimental results which support our categorisation of the costs and benefits of the proposed methods.