Reconstructing faulty measurements at a tailings treatment surge tank

Abstract

Abstract Measurement faults in processing plants can cause the performance of a process to deteriorate. Once the failure of a sensor is identified, it is possible to reconstruct the missing measurement using the measurements of other sensors. Plant operators or controllers can use the reconstructed measurement to make informed decisions. Although the theory to reconstruct faulty measurements is well-developed, various issues remain when applying the method in practice. In this study, a tailings treatment surge tank, which is a very simple process, is used to investigate issues surrounding measurement reconstruction using Principal Component Analysis. Different sets of faulty and correct sensors were created to investigate measurement reconstruction accuracy. The state observability of the surge tank model states was compared to the ability to reconstruct faulty measurements. It was found that the system does not necessarily need to be observable in terms of the available correct measurements for successful reconstruction. In addition, a fault in the measurement of the volume of slurry in the tank could not be reconstructed, even if it was the only faulty measurement. This indicates that the success of measurement reconstruction by Principal Component Analysis may depend on the dynamics of the process and the associated model.