The invention relates to an industrial process fault detection method based on correction type independent element analysis and Bayes probability fusion. A conventional fault detection method based on the correction type independent element analysis requires selection of a non-quadratic function to measure non-gaussianity. Different industrial process data or objects can cause a fact that enough experiential knowledge for guiding the selection of the non-quadratic function is hard to acquire in an actual application. By aiming at different non-quadratic functions, the normal data training of the industrial process is adopted to acquire different correction-type independent element models. The Bayes probability fusion method is used to integrate decision results of a plurality of fault detection models together to acquire a final probability type monitoring index. Compared with the prior art, the industrial process fault detection method is used to solve a model uncertainty problem caused by non-quadratic function diversity, and a plurality of model possibilities are fully considered, and the reliability of the fault detection models is enhanced to a great extent.