ARMA SIMULATOR FOR TESTING METHODS OF PROCESS DIAGNOSTIC
A methodology for global monitoring and on-line diagnosis of plants and process operations has been developed. It consists of a set of mathematical and statistical procedures enabling the process under study to be modelled, and the actual versus model-predicted behaviour to be contrasted. The study is subdivided into three sections: identification, rupture analysis, simulation. In the processes considered, the possible stationary states can be identified. Under some favourable physical conditions, e.g. soft changes, operational transients can be enclosed. Modelling, performed by a k-Multivariate Autoregressive Moving Average (k-ARMA), is suitable to fit time dependent stochastic systems, when linearity or quasi-linearity can be invoked. Suitable output functions of the process and its model can be contrasted and analysed on-line by statistical tests. Control card-like devices can be used to monitor system operations and report when and where the operations modelled no longer describe a healthy system. The set of these procedures has been compacted in the "Rupture Analyser" programme. A major theme of the work is to ascertain the practicality and sensitivity of the statistical rupture tests and their calibration in the cases under study. To this purpose an ARMA-generator simulating the foreseeable situations has been designed and implemented.
Bibliographic Reference: PAPER PRESENTED: COMPSTAT 9TH SYMPOSIUM ON COMPUTATIONAL STATISTICS, DUBROVNIK (YU), SEPT. 9-15, 1990 AVAILABLE FROM COMMISSION OF THE EUROPEAN COMMUNITIES, DG XIII-C-3, L-2920 LUXEMBOURG AS PAPER EN 35226 ORA
Availability: Can be ordered online
Record Number: 1989128086500 / Last updated on: 1990-11-01
Original language: EN
Available languages: en