This paper reports the effects of carbon fiber-reinforced polymer (CFRP) length on the failure process, pattern and crack propagation for a strengthened concrete beam with an initial notch. The experiments measuring load-bearing capacity for concrete beams with various CFRP lengths have been performed, wherein the crack opening displacements (COD) at the initial notch are also measured. The application of CFRP can significantly improve the load-bearing capacity, and the failure modes seem different with various CFRP lengths. The stress profiles in the concrete material around the crack tip, at the end of CFRP and at the interface between the concrete and CFRP are then calculated using the finite element method. The experiment measurements are validated by theoretical derivation and also support the finite element analysis. The results show that CFRP can significantly increase the ultimate load of the beam, while such an increase stops as the length reaches 0.15 m. It is also concluded that the CFRP length can influence the stress distribution at three critical stress regions for strengthened concrete beams. However, the optimum CFRP lengths vary with different critical stress regions. For the region around the crack tip, it is 0.15 m; for the region at the interface it is 0.25 m, and for the region at the end of CFRP, it is 0.30 m. In conclusion, the optimum CFRP length in this work is 0.30 m, at which CFRP strengthening is fully functioning, which thus provides a good reference for the retrofitting of buildings.

Wing shape is one of the main drivers of aircraft aerodynamic performance, so most aerodynamic shape optimization efforts have been focused solely on the wing. However, the performance of the full aircraft configuration must account for the fact that the aircraft needs to be trimmed. Thus, to realize the full benefit of aerodynamic shape optimization, one should optimize the wing shape while including the full configuration and a trim constraint. To evaluate the benefit of this approach, we perform the aerodynamic shape optimization of the Common Research Model wing-body-tail configuration using gradient-based optimization with a Reynolds-averaged Navier-Stokes (RANS) model that includes a discrete adjoint implementation. We investigate the aerodynamic shape optimization of the wing with a trim constraint that is satisfied by rotating the horizontal tail. We optimize the same wing-body configuration without the tail, but with an added trim drag penalty based on a surrogate model we created. We also conduct the simultaneous aerodynamic shape optimization of both the wing and the horizontal-tail to investigate the benefit of optimizing the tail shape. The design variables consist of 816 wing shape variables and 144 horizontal tail shape variables, as well as the tail rotation angle and the aircraft angle of attack. The drag coefficient is minimized subject to lift and trim constraints. In addition, 1000 geometric thickness constraints and a wing volume constraint are enforced to guarantee that the wing thickness and volume values are no lower than those of the baseline geometry. We found that considering the trim during optimization is a better approach than using a fixed wing moment constraint. We also showed that the trim drag surrogate model we created can be used to yield a design whose drag coefficient is within 1.2 counts of that of an optimization where trim is satisfied with a rotating tail. However, we recommend the simultaneous optimization of wing and tail rotation to obtain the best possible performance.

Characterizing compute kernel execution behavior on GPUs for efficient task scheduling is a non-trivial task. We address this with a simple model enabling portable and fast predictions among different GPUs using only hardware-independent features. This model is built based on random forests using 189 individual compute kernels from benchmarks such as Parboil, Rodinia, Polybench-GPU, and SHOC. Evaluation of the model performance using cross-validation yields a median Mean Average Percentage Error (MAPE) of 8.86–52.0% for time and 1.84–2.94% for power prediction across five different GPUs, while latency for a single prediction varies between 15 and 108 ms.

No abstract

No abstract