Machinery Condition Monitoring by Inverse Filtering and Statistical Analysis

Data used for machinery condition monitoring contains mainly the same information as that obtained under normal operation conditions. The traditional practice of feature extraction, which uses such data directly, suffers from low signal-to-noise ratio. This paper presents a method that uses an inverse filter to separate the information contents of the data, so that the feature extraction can be done by statistical analysis algorithms, which would otherwise be difficult. It is shown that the inverse filtering process is equivalent to that of prediction error estimation based on a signal model in the form of an autoregressive moving-average (ARMA) model. The construction of the inverse filter can therefore be carried out by ARMA modeling. An application example of this method for the monitoring of a paper handling system is also given.