A Lightweight High-Impact Acceleration State Reconstruction Method for Multibody Dynamic Systems by an Extended Kalman Filter- Aided Time Neural Network

The complex mechanical transfers in multibody dynamical systems lead to large deviations of the measured signals of accelerometers and other sensors from their real state quantities, which results in difficulties in state identification and decision-making for mechatronic equipment. A novel state reconstruction method of extended Kalman filter (EKF)-aided time neural network is proposed for multibody dynamical systems in this article. First, we designed a least-squares dual extend Kalman filter (DLEKF) method for multibody dynamic systems with parameter uncertainties, which includes a system state filter and structural parameter estimator. Then, we design a multidimensional short time-series lightweight gated recurrent unit (GRU) neural network with DLEKF, which realize state reconstruction of unmeasurable subsystems by relying on less input data and computational resources. Finally, the effectiveness of the proposed method is verified via simulations and experiments. Compared with the extend Kalman filter (KF) and long time-series neural network methods, the mean square error (mse) of the proposed method in this article is reduced by 79.9% and 67.7%. Furthermore, we take an interpretable analysis to argue the superiority of the proposed method and prove that the proposed method has good application prospects.