Fault detection in dynamic processes using a simplified monitoring-specific CVA state space modelling approach

State space models have been successfully used for the modelling, control and monitoring of dynamic processes with several different approaches employed to derive the state variables of the model. Typically, state-space canonical variate analysis (CVA) modelling requires the estimation of five matrices to fully parameterize the model. This paper proposes a simpler CVA state space model defined by three matrices for the specific purpose of process monitoring. A modified definition of the past vector of inputs and output is proposed in order to facilitate efficient estimation of a reduced set of state space matrices. A sequential procedure for accurate selection of the model state vector dimension is also proposed. The proposed method is applied to the benchmark Tennessee Eastman process and the results show that the proposed method gives comparable and in some cases even better performance than the established CVA state space monitoring methods.

► A simplified CVA based state space model for the specific purpose of process monitoring is proposed. ► Using a modified definition of the past vector of inputs and outputs the CVA model has 3 instead of 5 parameter matrices. ► A sequential procedure for the accurate selection of the state vector dimension is proposed. ► The proposed method is successfully applied to the simulated Tennessee Eastman process.