In order to get a more accurate finite element analysis model of the bolt-ball shell for health monitoring, the element in adjustable stiffness model is used to reveal the semi-rigid characters of the node. Then, the neural network technology is introduced, by forming network input parameter CPFM with limited measuring point information. A new method of the recognition of the rigid factor, in consideration of the semi-rigid character of the joint reasonably, is put forward by the combination of the elements. A single-layer latticed cylindrical shell with 157 nodes and 414 elements experiment model was used for shaking table test. Based on the basic model and measured modal data, a finite element model updating is done. The result shows that the true dynamic characters of the shell structure can be reflected better, at the same time the neural network can be simplified by applying step-by-step correction algorithm. The result shows that the method can be used for model updating of the bolt-ball shell and certain value in engineering practice.