A real-time fault diagnosis method for inertia device based on variational mode decomposition is proposed to ease the difficulty of fault recognition for slow-varying output. The correlative modal components related to output data of inertia device are calculated by variational mode decomposition method firstly, subsequently the number of modal components is confirmed adaptively with frequency discrimination, then the mapping strategy from modal components to fault eigenvector is defined, trustworthy fault diagnostic capability is built with fault classifier based on probabilistic neural network at last. Multiple fault diagnosis simulation is carried out with average correct rate at least 80%. Considering there is no need for the operating principle of the inertia device to participate in the diagnosis process, the fault diagnosis method can be applied in launch vehicles in order to supplement approaches for health state assessment of inertia devices.