Investigating the problem of misdiagnosis in model-based fault diagnosis

This paper deals with one partly unconscious property of the model-based diagnosis. It discusses occasional contradictions between diagnoses that are logically correct but, in fact, are not consistent with the physical state of the system being diagnosed. This property is studied and discussed based on the analysis of diagnoses generated by four selected approaches using binary and trivalent diagnostic signals. The authors attribute the reasons for this inconsistency to the effect of compensation of fault impacts. The analysis and simulation studies carried out confirmed this assumption. To address this problem, new definitions of diagnoses have been proposed that reflect the different degrees to which diagnoses relate to the actual physical state of the system being diagnosed. In this context, several new metrics for assessing the quality of diagnoses have also been proposed. It is pointed out that, from the utilitarian point of view, only those diagnoses that are logically consistent and have the attribute of physicality are valuable. The problem of misdiagnosis was illustrated on an example of a two-tank system.