According to the characteristics and requirements of fault diagnosis in process systems and based on the previous work

this paper puts forward a Sigmoidal - Quadratic Basis Function Network (SQBFN) which can improve the bounding capability of fault space. This netv/ork first divides the input space by quadratic hypersurfaces and then exactly describes the regions of faults in fault space through complicated decision space formed by the superposition of those hypersurfaces. SQBFN is of powerful classification

good flexibility and may form the required closed or semi - closed decision spaces avoiding arbitrarily dividing blank spaces. This paper gives an introduction to SQBFN and verifies its effectiveness by two examples. The results of theoretical analysis and researches on examples indicate that the bounding capability of SQBFN is superior to other networks and at the same time its structure is not complicated and even more simple.