Canonical correlation analysis(CCA)can be implemented to monitor a linear process when the input-output relationship is explicitly existing.However,the conventional CCA method may not be well-suited to monitor and detect the fault in multimode processes.In this paper,a novel approach is proposed based on CCA for addressing this problem.The multiple model is assumed to follow a Gaussian mixture model.With the help of EM algorithm,firstly,the parameters of mixture model could be estimated.Then a Bayesian inference based test index is developed to detect the faults.The validity and effectiveness of the proposed monitoring approach are illustrated through two applications.The first one is a simple multivariate linear system.The second one is a simulated continuous stirred tank heater(CSTH)benchmark process.The comparison of monitoring results in the two examples demonstrates that the proposed method is superior to the conventional CCA-based approach.