Online monitoring method for geometric quality in fuselage panel riveting based on MBCNN-ViT

To enhance the level of intelligent inspection in automatic drilling and riveting in aircraft manufacturing, this paper presents a geometrical quality prediction method for rivet pressing based on hybrid datasets and MBCNN-ViT. Firstly, simulations are conducted based on the real dataset to generate the simulated dataset. Then, a hybrid dataset combining real and simulated data of rivet pressing. Gramian angular fields (GAF) was employed to preprocess the data. By integrating MBConv convolutional blocks and ViTransformer (ViT) encoding layers, the MBCNN-ViT prediction model is constructed. Through training and experimental validation, the proposed method achieves predictions for heading height and diameter with an error margin of ±0.02 mm. Deploying the proposed method on production equipment enables online monitoring of the riveting geometrical quality, demonstrating the effectiveness and applicability of the model presented in this paper.