Dynamic model identification method of manipulators for inside DEMO engineering

•A method to construct the dynamics model of manipulators with unknown parts is proposed.•The BP neural network is used to identify and substitute the unknowns of the dynamic system.•A modified Levenberg-Marquardt algorithm is developed to train the BP neural network using the system front-end input and output data.•Dynamic model of a parallel kinematic mechanism with unknown friction models is constructed successfully. In the inside engineering of DEMO, the robotic machines and manipulators are envisaged to be widely employed, which often have to deal with the demanding working conditions. The construction of the dynamic model of the robot and manipulator can be incorporated into the control system design to gain the adaptive control performance. A method of constructing the dynamic model with the unknown parts is proposed. The method can identify the unknown parts of the dynamic system by incorporating a BP neural network that will eventually substitute the unknown parts in the system dynamics after the well training. A modified Levenberg-Marquardt algorithm is developed for the training of BP neural network, which can back propagate the errors between entire actual system and the constructed dynamic model into the training process of the neural network. An example of constructing the dynamic model for a general Stewart structure mechanism is presented, in which the unknown parts, as well the entire dynamics are successfully identified. The proposed method can be extrapolated to the dynamics modeling of the blanket transporter in DEMO design, as well as the other general manipulators.