Abstract: Because the sensors have noises and, the failure discrimination rule is sometimes unreliable, the fault detection result of a machine is not always correct. A good fault diagnosis method should contain redundancy among different test points and instants to increase the fault isolation rate. The dynamic multiple fault diagnosis (DMFD) method is shown to have a good diagnostic performance in testing unreliable conditions. But when the system is large in scale, the iteration would be unnecessary, and it does not always reach the optimal value. In this paper, the principles of Lagrange relaxation and Viterbi algorithm are analyzed first, then a method is proposed to make use of the relationship between faults and test results to estimate the Lagrange multiplier. By leaving out the iteration process, the computation time is reduced and the real-time performance of the algorithm is enhanced. Simulation results show that the new algorithm gives much faster calculation speed with a higher fault isolation rate.