Abstarct: Surge is a serious harmful instability in compressors that affects the entire compression system. For quick and accurate surge suppressing, the real and healthy values of fluid flow, pressure ratio and speed of compressor sensors as key variables should be accessible any time. Currently, there is not any sensor fault diagnosis (FD) in the compressor. Statistical and field studies in the gas compressor station show that compressors shut down frequently due to absence of FD. To prevent performance deterioration, major collapses, for instance surge phenomenon in the compressor as a sophisticated, expensive and safety critical system, advanced technologies for early FD and control must be incorporated into engineering designs. However, no considerable and impressive work has previously been reported on the use of sensor FD within compressor surge suppressing system. In this thesis, the neural networks, Dynamic neural network baxsed on robust identification and Support vector regression are used as model free technique tools to diagnosis of compressor sensor faults. The type, amplitutde and behaviour of fault can be identified by proposal residual evaluation block. Subspace identification method is used for generating of Hurwithz matrix in the dynamic neural network baxsed on robust identification. Groebner basis and elimination theory is also used as a model baxsed method for fault diagnosis. The results of fault diagnosis systems are used in virtual sensor method as a fault tolerant control system.