Flexibility prediction of thin-walled parts based on finite element method and K-K-CNN hybrid model

Elastic deformation in thin-walled parts during machining is affected by the coupling of force and flexibility. Obtaining flexibility information along machining tool paths is crucial for online monitoring of this deformation. However, the current finite element method (FEM) is limited by mesh nodes, hampering its ability to accurately determine flexibility along tool paths. To overcome this limitation, this paper proposes a method that combines FEM with the surrogate model to predict flexibility accurately at any position on thin-walled parts’ surfaces. The surrogate model is the hybrid model K-K-CNN based on two K-nearest neighbor (KNN-KNN) algorithms and a convolutional neural network (CNN) model. Initially, an initial dataset containing positions and flexibility of mesh nodes is generated automatically through secondary development of ABAQUS. Then, the K-K-CNN hybrid model is introduced and trained on this dataset to calculate flexibility accurately at any position on the surface of thin-walled parts. The hybrid model employs a CNN to address the nonlinear spatial correlation issue in flexibility prediction. Moreover, the hybrid model incorporates two KNN algorithms to alleviate the overfitting challenge stemming from the straightforward input features and extensive dataset size. In comparison to traditional deep learning models, the K-K-CNN hybrid model presents notable benefits in predicting flexibility for complex thin-walled parts at any given position, which affirms its robustness and accuracy. The proposed prediction method for flexibility can provide high-quality data-driven information for monitoring the elastic deformation of thin-walled parts.