Force Decoupling for Assembly Interface of Large Aircraft Structures by Neural Networks

In this paper, a neural-network-based method for decoupling the contact forces between assembly interfaces and motion actuators is explored, aiming at projecting the actual forces on assembly interfaces during the assembly of large aircraft structures. Due to the spatial limitation and geometrical complexity of the assembly interface, it is difficult to directly measure the forces on the assembly interface during pose adjustment and aircraft cabin assembly by traditional methods. Therefore, it is proposed to decouple the assembly surface forces through the relationship model between the data from three-dimensional force sensors and the forces at the assembly interface to ensure the effectiveness of force control and monitoring. Aiming at the errors existing in the force model of the traditional center-of-gravity method, this paper proposes a neural network to construct the force decoupling model, gate-separation multilayer perceptron, and uses the feature-separated gating training to design the input–output relationship to improve the force prediction accuracy. During the training process, the model stability and generalization ability are ensured by the mean square error loss function, network structure screening, and hyperparameter tuning. The model can detect the force condition of the assembly interface in real time, which provides the basis for the force control in the flexible assembly process. Experimental results show that the method can effectively optimize the force prediction accuracy between assembly interfaces during pose adjustment.