Fault isolation method for nonstationary industrial processes

Abstract

It is very important to isolate the faulty variables after a fault is detected. However, it is challenging to isolate the faulty variables due to the nonstationarity which widely exists in the industry processes. The statistical properties of nonstationary process variables are time-variant, i.e., these variables are time-dependent. This paper proposes an effective faulty variable isolation method for the nonstationary industrial processes using the cointegration method. In the nonstationary industrial processes, not all the variables are nonstationary. The nonstationary variables should be distinguished from the stationary ones. Then, the nonstationary variables are used to build the cointegration model to describe the long-run equilibrium relation among those nonstationary variables. Finally, the least absolute shrinkage and selection operator method is integrated to the cointegration model for selecting the faulty variables that are mainly responsible for the fault. The proposed faulty variable isolation method can deal with the nonstationary issue in the industrial processes and isolate multiple faulty variables simultaneously. Its feasibility and performance are illustrated with a real industrial process of the thermal power plant.