Fault diagnosis based on qualitative/quantitative process knowledge

Recent advances in knowledge engineering have led to develop the qualitative (deep) model based diagnostic systems. As process knowledge accumulated, however, the diagnostic system remains strictly qualitative. This limits the usefulness of such systems. In this work, a framework is developed for integrating quantitative process knowledge into the qualitative model. Once quantitative process information, e.g., steady-state gains, is available, it can be incorporated into the simplest qualitative process model called the signed directed graph. The quantitative process knowledge is described in terms of membership functions of fuzzy set theory. According to the measurement pattern, the truth values of a hypothesis (e.g., a fault origin) can be calculated based on the fuzzy logic. Consequently, the diagnostic resolution can be improved significantly. Furthermore, the proposed method becomes a strictly qualitative diagnostic system, if no quantitative information is available. A chemical reactor example illustrates the design and performance of the qualitative/quantitative model-based diagnostic system. The proposed approach can also be extended to the multiple-fault situations in a straightforward manner.