Prediction of the dynamic performance for the deployable mechanism in assembly based on optimized neural network

Abstract Since manufacturing errors significantly affect the dynamic performance of the deployable mechanism, the length of links must be adjusted in assembly. To ensure the assembly quality and shorten the adjustment period, this paper presents an efficient assembly optimizer with machine learning for the deployable mechanism. Considering the assembly dimensions, a numerical iterative algorithm is firstly proposed to analyze ultimately deployed angles of joints after link adjustment. For more effective inference, an agent model with neural network is trained to predict the dynamic performance. The training datasets are obtained after calculating amounts of locked angles for joints in different dimensional errors, and a BP network with two hidden layers is constructed based on optimal brain surgeon. Experiments demonstrate that our method can accurately predict whether the adjusted mechanism meets the requirement of assembly quality in about 0.3 s.