A comprehensive computational micromechanics has been presented to predict the transverse mechanical properties and failure envelop of bi-axial stress in a 22-33 plane at high temperatures. First, a paraboloidal elasto-plastic damage model considering the temperature effect of epoxy has been employed. Then, in order to rapidly create the random fiber distribution of the representative volume element (RVE), a simple and effective algorithm of fiber placement and an auto modeling and calculating python script have been developed, where the thickness size effect of RVEs has been discussed. Finally, the transverse tension/compression and out-of-plane shear properties and the corresponding failure envelop of RVEs are analyzed systematically by means of finite element simulations. The results show that the in-plane shear strength exists strongly thickness size effect, and high temperature can significantly weaken the transverse strength, especially for transverse compression, and the temperature effect of modulus can be depicted by a specific Gaussian function. In addition, Hashin and Tsai-wu criteria may be more appropriate to predict the bi-axial failure envelop of ${\sigma }_{2}2-{\tau }_{23}$ at high temperature, and Tsai-Wu criterion is better and more suitable than Hashin and Catalanotti criteria for considering bi-axial compression.

The longitudinal compressive behavior of the three-dimensional four-step braided composites after cyclic hygrothermal aging was investigated using a split Hopkinson pressure bar (SHPB) apparatus under high strain rates (1100~1250 s⁻¹, 1400~1600 s⁻¹, 1700~1850 s⁻¹, respectively). The SEM micrographs were examined to the damage evolution of the composites after cyclic hygrothermal aging. A high-speed camera was employed to capture the progressive damage process for the composites. The results indicate that the saturated moisture absorption of the composites was not reached during the whole 210 cyclic hygrothermal aging days. The composites mainly underwent epoxy hydrolysis and interfaces debonding during continuous cyclic hygrothermal aging time. The peak stress of the composites still behaved as a strain rate effect after different cyclic hygrothermal aging days, but the dynamic stiffness modulus clearly had no specific regularity. In addition, the peak stress and the dynamic stiffness modulus of the composites after 210 cyclic hygrothermal aging days almost decreased by half when subjected to longitudinal compression.

Could an artificial intelligence (AI) system have recognised material fatigue before engineers had named and systematically investigated it? This counterfactual question provides a demanding test of the distinction between successful prediction and physical explanation. Here, we construct a historically bounded thought experiment using early writings on strength and elasticity, while recognising an important complication: Tredgold [1] had already described fracture under repeated flexure in 1822, decades before the later systematic development of fatigue research. Any apparent rediscovery of fatigue by a present-day language model therefore cannot, by itself, be regarded as independent scientific discovery. We instead formulate an auditable test in which an AI system must recognise an anomalous failure, distinguish among competing explanations, propose experiments capable of discriminating between them, and make prospective predictions without access to the later conceptual solution. The purpose is not to claim that contemporary AI has failed such a test, but to clarify what evidence would be required before attributing genuine physical discovery to an AI system. The broader implication is that computational prediction can substantially extend scientific capability, whereas experimentally accountable interpretation remains a central component of scientific understanding.