To improve the convergence rate of neural networks and overcome the shortcoming of easily falling into local extreme, the coordination of dynamic acceleration constant and inertia weight called WCPSO algorithm is proposed to optimize network parameters. This algorithm improves the WPSO to realize linear evolution of the dynamic acceleration constant. In this paper, WCPSO was applied to marine diesel fault diagnosis model, and the simulation results show that the optimized diagnosis model is more prepared and the diagnosis speed is faster.