Abstract: In order to handle the problem of fault diagnosis for industrial processes, an improved fault detection method was proposed based on canonical variable analysis (CVA) and independent component analysis (ICA). At the same time, combined with continuous string matching (CSM), a new fault diagnosis method based on the library of complete faults was proposed. First, the CVA algorithm was used to calculate the canonical variable of the data, and then, the ICA algorithm was used to decompose the canonical variable. Finally, the CSM algorithm was used to diagnose the faults. A case study of Tennessee Eastman (TE) process showed that the proposed algorithm is feasible and efficient.