In order to enhance the fault tolerance ability of the fault diagnosis model,information correction is thus adopted based on the maximum likelihood decoding theory.

On the one band, the maximum likelihood decoding of linear block codes in the soft-decision condition is proved to be equivalent to making a neural network converge to a globe stable state of the energy function.

In this paper the function of error-corecting decoding of competitive learning is approached,the response,at equilibrium,of this neural network is shown to be equivalent to performing maximum likelihood decoding in a binary linear block error-correcting code on the basis of analysing the principle of competitive learning.