Conference article

Industrial process failures can be often seen as a variance increase in a measured process variable. The objective of this research was to investigate if stochastic Autoregressive Moving Average, abbreviated as ARMA, and Generalized Autoregressive Conditionally Heteroscedastic, abbreviated as GARCH, time series modelling are feasible methods for the reliable detection of gradually increasing variance in the process variable. A case study was conducted for the reliable detection of increased pressure variance that indicates a harmful air leakage in a vacuum chamber in a paper machine. Variance in the chamber pressure was artificially gradually increased, a combined ARMA+GARCH time series model was fitted to it and the variance vector was determined. An abnormally high variance was detected from the variance vector using a specified detection limit and detection sensitivity. According to the simulation results, by controlling the variance vector extracted from the combined ARMA+GARCH time series model, a very slight variance increase in the process variable can be detected more reliably than detecting it from the moving variance vector computed directly from the process variable.

condition monitoring, time series analysis, SPC