Fault detection in boilers using canonical variate analysis

A software based system was developed that monitors industrial boilers on-line for the purpose to detect and identify fault conditions. Dynamic models are used to predict an expected set of process conditions, which form the baseline for evaluation of process conditions. By continuously monitoring the deviations between the baseline and actual observations, insight in the presence of certain faults can be obtained. An online expert system is used to draw conclusions from the observed prediction errors. The dynamic models are based on state space models which are generated by system identification using canonical variate analysis (CVA). The objectives of the system, the approach to the fault detection process and experience with a commercially available system identification tool are discussed