Fault detection and isolation using local models, comparison with unknown input observers

This paper describes a model-based fault isolation method where the model of the whole fault-free system is partitioned into multiple local models. By feeding the local models with the available measurements, prediction residuals can be calculated. They are shown to have strong decoupling properties. The prediction residuals are reformulated in the unknown input observer (UIO) framework in order to compare the two approaches. The proposed residual generation method is applied to the diagnosis of a DC motor current control loop.