In modern manufacturing, thin-walled frame beam parts are widely used in the various industries due to their lightweight and high-strength characteristics. However, their low stiffness leads to a significant machining deformation caused by the release of internal stress field (residual stress generated during forming processes) of the blank, adversely affecting the precision and performance of final parts. To address this issue, an inverse-construction method for internal stress field that combines finite element simulation with deep learning is proposed in this study, overcoming the limitations of traditional measurement techniques such as destructiveness, high costs and data insufficiency. Taking aluminum alloy as the research object, a three-dimensional finite element model of the thin-walled structure is constructed. The rough machining process is simulated to obtain a mapping dataset between the initial internal stress field and the rough machining deformation field. A deep learning network model is then developed with the rough machining deformation field as input and the initial internal stress field as output, enabling the inverse-construction of the initial internal stress field based on the rough machining deformation field. This study enables a potential of deformation field prediction of successive machining processes based on deep learning technology, demonstrating the feasibility for a new approach of smart manufacturing, which can lay a foundation for digital twin technology of thin-walled part machining.