Application of Canonical Variate Analysis for fault detection and performance control

The detection and diagnosis of faults in industrial processes can help to reduce maintenance costs and improve the safety of plant operations. In order to avoid inefficient operation of faulty equipment and minimize the number of unnecessary shutdowns it is important for the operators to have reliable information about the current impact of the fault in the process performance and the future evolution of the fault. This can be a complex topic when dealing with systems working under varying operating conditions. Canonical Variate Analysis (CVA) is a multivariate algorithm for process monitoring which has the ability to capture process dynamics more efficiently than other similar methods. The aim of this work is to demonstrate the capabiliity of CVA to extract reliable information about the process condition and the effects of the faults in the system performance using experimental data. The data sets were acquired from a large experimental multiphase flow facility representing a real small-scale multiphase flow separation process. Different faults were introduced in the system to assess the performance of CVA in terms of fault detection and diagnosis, and also to model the system behaviour under normal and faulty conditions. The results suggest that CVA is a very reliable tool for fault detection and diagnosis, as well as for the identification of the system behaviour.