The process monitoring and diagnosis in assembly process is important. Multivariate T2 control charts are applied to detect the mean shift and interaction change in the assembly process. However, T2 charts can not identify the root cause of the change. The traditional MTY method for T2 signal decomposition is computationally expensive, especially when the dimension of the variables is high. A new approach based on Bayesian network to identify the significant cause of T2 signals is proposed in this paper. The headlamp bracket case is used to illustrate the overall procedure. And the effectiveness of the proposed approach is evaluated.