A Multi-View Feature-Based Retrieval Method for On-Machine Measurement Model Reuse

Abstract

In the design of on-machine measurement (OMM) schemes for complex parts, the involvement of diverse geometric features, physical environments, and hardware/software systems makes the process intricate and inefficient. To address the high dependency on manual experience and the low efficiency in current OMM scheme design, this paper proposes a multi-view 3D model retrieval method aimed at facilitating the reuse of existing measurement solutions. Specifically, multiple view images of the same model are processed using a multi-view convolutional neural network (ResNet-18) to extract view-specific features. A self-attention mechanism is introduced to enhance the network’s focus on informative views and critical features. These features are then fused into a global feature descriptor through a long short-term memory (LSTM) network. Finally, similarity between the query model and models in the feature database is computed, and retrieval is performed by ranking similarity scores in descending order. Based on the retrieved models, corresponding measurement schemes can be adjusted and reused efficiently. Experimental results demonstrate that the proposed method outperforms the MVRNN approach in terms of Precision and Recall, and improves the mean Average Precision (mAP) by 5%. The method effectively enhances the efficiency of OMM scheme design and shows strong potential for practical engineering applications.