Data-driven diagnosis with ambiguous hypotheses in historical data: A generalized Dempter-Shafer approach

This work addresses the problem of diagnosis with ambiguous hypotheses in the historical data, which has its interest in the fields of fault and control-loop diagnosis. Our earlier work showed that Dempster-Shafer theory was adequate to formulate direct probability estimates with data from ambiguous hypotheses, but was inadequate to represent likelihood estimates in similar circumstances. This work extends Dempster-Shafer theory so that it can formulate a generalized Basic Belief Assignment (BBA) that represents the linear approximation of the likelihood, and use a generalized rule to combine generalized BBAs. This method was applied to a simulated industrial solids handling facility.