The invention discloses an industrial process fault detection method, and the method comprises the steps: obtaining the operation data of a to-be-detected industrial process, carrying out the processing of the operation data through employing a preset sliding window, and obtaining operation data matrixes under different windows; calculating the eigenvalue and eigenvector of a covariance matrix of the first operation data matrix, and determining the eigenvalues and eigenvectors of the covariance matrixes of other operation data matrixes in an iterative manner; calculating the detection indexes of all operation data matrixes according to the eigenvalues and eigenvectors of the covariance matrixes of all operation data matrixes, comparing all detection indexes with a preset detection index threshold value, and judging whether there is a fault in the to-be-detected industrial process or not according to the comparison result. Compared with a blast furnace abnormity detection method based on an expert system, the method does not historical fault information, avoids a tedious rule training process, is easier to implement, and is higher in efficiency.