On robust online identification of industrial systems

Abstract

In the industry, due to transport phenomena and non-linearities, high order linear models are often used to fit the process behaviour. In several applications, a first order model with a delay can give a satisfying result when compared to the actual process. This paper focuses on the identification of a continuous time first order plus dead time (FOPDT) process. In this context, identification of process dynamics is important because it enables the process engineer, to quickly design a control strategy that meets customer's requirements. In this paper an instrumental variable recursive least squares, with a variable forgetting factor (VFF-IVRLS) algorithm is proposed. The online implementation is done by adding robust start and stop conditions. A Schneider PLC is used to apply step tests in order to identify the system response. The robustness of the proposed methodology is discussed and tested with real industrial data. The data is collected by the sensors of the system.