Three-dimensional crack growth prediction for attachment lugs based on physics-knowledge neural network

As a typical structure that connects important components in aircraft, lug joints will inevitably experience fatigue damage during service, which will pose a significant threat to aircraft safety. However, due to the complexity of the geometric structure and the three-dimensional (3D) crack in the real world, it is challenging to conduct damage tolerance analysis for different lug structures. In this paper, a set of computational procedures for 3D crack growth of attachment lugs based on physics-knowledge neural networks is proposed. The stress intensity factor database is provided by finite element method and physics-knowledge neural networks, and the crack growth analysis is based on 3D fatigue fracture theory. The growth behaviors of through-thickness cracks, corner cracks, surface cracks of straight lugs, and corner cracks of tapered lugs under constant amplitude or random spectrum loads are predicted and compared against available experimental results in the literature. The lug materials include 4340 steel, 7075-T651 aluminum, and polymethyl methacrylate. For through-thickness crack straight lugs, the effect of interference fit on the crack life is analyzed. For quarter elliptical corner crack tapered lugs, the crack growth under different loading directions and crack positions is investigated. The predicted results of different lug models are in good agreement with the experimental results, which verifies the accuracy of the computational procedures in this paper. (Figure presented.)