In this paper, a non-intrusive pressure measurement scheme based on particle image velocimetry (PIV) is presented for the complex supersonic flows with intense shock systems, by elaborately combining the MacCormack method, the streamline-based method, and the spatial integration in conservative form. According to the detailed analyses of flow structures, the pressure fields are well reconstructed by the proposed scheme for the two typical shock-wave/boundary-layer interactions containing regular and Mach reflections, which are induced by the relatively strong oblique shock waves generated by the wedges of 21° and 17° in the freestreams of Mach 2.5 and 2.0, respectively. Based on the theoretical solutions by oblique shock relationship, free interaction theory, and shock polar analysis, this pressure reconstruction scheme is completely validated to effectively suppress the propagation of PIV velocity error to the pressure field and the accumulation of reconstructed pressure error behind the strong shock wave. Compared with the literature presently, this work would be the most challenging application of PIV-based pressure measurement to such complex supersonic flows with intense shock reflections, large oscillations, wide speed ranges, and various compressible flow structures. These good results could confirm the feasibility and high accuracy of the proposed reconstruction scheme and may greatly promote its applications in academic research and engineering test for supersonic flows in the future.

AI applications powered by deep learning models are increasingly run natively at edge. A deployed model not only encounters continuously evolving input distributions (domains) but also faces adversarial attacks from third-party. This necessitates adapting the model to shifting domains to maintain high natural accuracy, while avoiding degrading the model's robust accuracy. However, existing domain adaptation and adversarial attack preventation techniques often have conflicting optimization objectives and they rely on time-consuming training process. This paper presents RobustDA, an on-device lightweight approach that co-optimizes natural and robust accuracies in model retraining. It uses a set of low-rank adapters to retain all learned domains' knowledge with small overheads. In each model retraining, RobustDA constructs an adapter to separate domain-related and robust-related model parameters to avoid their conflicts in updating. Based on the retained knowledge, it quickly generates adversarial examples with high-quality pseudo-labels and uses them to accelerate the retraining process. We demonstrate that, comparing against 14 state-of-the-art DA techniques under 7 prevalent adversarial attacks on edge devices, the proposed co-optimization approach improves natural and robust accuracies by 6.34% and 11.41% simultaneously. Under the same accuracy, RobustDA also speeds up the retraining process by 4.09x.

Hexagonal boron nitride (BN) is often used as filler to improve the thermal conductivity of polymer matrix due to its high thermal conductivity. However, previously reported BN-based composites always have a high in-plane thermal conductivity, which is not beneficial for vertical heat dissipation. In addition, high BN content results in the deterioration of the mechanical properties. Here, we report a feasible method to prepare a BN/silicone rubber (SiR) composite with oriented BN in organosilicon matrix via a vacuum-assisted self-assembly technique. The BN/SiR composite displays a 1270% higher (2.74 W/(m·K)) thermal conductivity than that of neat organosilicon matrix (0.20 W/(m·K)). The oriented BN nanosheets increase the polymer's adhesive force and exhibit excellent compression cycle performance. In turn, these features support its superiority as thermal interface material in the light-emitting diode chips heat dissipation application.