Neural networks for fault diagnosis in spacecraft

The BP algorithm proposed for neural networks is discussed, which made the neural networks for practical use since the neural networks has not been applied properly because the user failed to stress the pretreatment of input data to the neural networks. An improved Cauchy method to normalize input data in the light of the characters of spacecraft and methods of constructing normalization functions are proposed. The comparison results of these methods show that normalization is the key reason on the correctness of faults diagnosis in spacecraft.