A large fraction of the algorithms that support predictive scheduling of maintenance essentially are anomaly detection algorithms. A model is built of the normal range of the measurable signals. When the system departs from the normal range, an alarm is sounded. Various statistical technologies have been used to model the normal signals such as neural networks , principal components analysis , or multivariate state estimation.
Anomaly detection techniques are very useful for device monitoring, and in fact are the only possibility in applications where identical devices similarly used nonetheless produce very different measurements.
We have extensive experience in applying state of the art algorithms to the problem of impending failure detection. The figure shows the time evolution of 18 sensors of one of our former client's devices. Our state-of-the-art signal processing algorithms were able to isolate a slowly developing fault which -in this case- is hidden to the naked eye.