Aiming at the nonlinear system fault d iagnosis in multi-sensor observations,a novel sequential maximum likelihood rat io test for fault diagnosis based on multi-sensor particle weight optimization is proposed.Firstly,to improve the adverse effects on the stability of particle weight measurement caused by ra ndom noise,the particle weight optimization strategy is designed by multi-source information fusion technology to fully extract and exploit the redundancy and complementary information from multi-sensor observations.Its im plementation principle is to improve the reliability and stability of particle weight by the decline of particle weig ht variance,and to promote the estimation precision.Secondly,combined with the sequential probability ratio test and the i nteracting multi-model,a novel online sequential maximum likelihood ratio method based on residual test is presented.I n addition,considering the decline of computation complexity in the combination process of particle filter and interacting multi- model,the input interaction step and the output step are reasonably simplified in the construction of algorithm.The theo retical analysis and experimental results show the feasibility and efficiency of the proposed algorithm.