Deep and Shallow Knowledge in Fault Diagnosis
Diagnostic reasoning is fundamentally different from reasoning used in modelling or control: last is deductive (from causes to effects) while first is abductive (from effects to causes). Fault diagnosis in real complex systems is difficult due to multiple effects-to-causes relations and to various running contexts. In deterministic approaches deep knowledge is used to find ”explanations” for effects in the target system (impractical when modelling burden appear), in softcomputing approaches shallow knowledge from experiments is used to links effects to causes (unrealistic for running real installations). The paper proposes a way to combine shallow knowledge and deep knowledge on conductive flow systems at faults, and offers a general approach for diagnostic problem solving.
Unable to display preview. Download preview PDF.
- 1.Ariton V.: Fault diagnosis connectionist approach for multifunctional conductive flow systems. PhD dissertation, Galati (1999) Google Scholar
- 2.Ariton, V., Bumbaru, S.: Fault Diagnosis in Conductive Flow Systems using Productive Neural Networks. In: The 12th International Conference on Control Systems and Computer Science CSCS12, Bucureşti, pp. 125–130 (1999)Google Scholar
- 3.Cellier, F.E.: Modeling from Physical Principles. In: Levine, W.S. (ed.) The Control Handbook, pp. 98–108. CRC Press, Boca Raton (1995)Google Scholar
- 4.Larsson, J.E.: Knowledge-based methods for control systems, PhD Thesis Dissertation, Lund, Sweden (1992) Google Scholar
- 5.Mosterman, P.J., Kapadia, R., Biswas, G.: Using bond graphs for diagnosis of dynamical physical systems. In: Sixth Intl. Conference on Principles of Diagnosis, Goslar, Germany, pp. 81–85 (1995)Google Scholar