Improving the resolution and stability of diagnostic systems

The paper addresses the problem of distinguishing between faults with the same fault signatures. It is shown that additional improvements in terms of resolution can be achieved if information about the sensitivity of the residuals with respect to faults is taken into account. When two fault signatures are qualitatively identical but have significantly different sensitivity terms, additional discrimination can be achieved by using the ratios of the underlying residuals. This extra isolability feature simplifies reasoning in the isolation stage. However, to achieve better overall performance, the diagnostic results should be stable, i.e. should not vary substantially under different operating conditions. Improved diagnostic stability is obtained by the approximate reasoning approach which relies on the transferable belief model. An outline of the ideas is given and some practical results obtained on a DC motor are provided.