Search for: All records

Abstract The objective of this work is to reduce the cost of performing model-based sensitivity analysis for ultrasonic nondestructive testing systems by replacing the accurate physics-based model with machine learning (ML) algorithms and quickly compute Sobol’ indices. The ML algorithms considered in this work are neural networks (NNs), convolutional NN (CNNs), and deep Gaussian processes (DGPs). The performance of these algorithms is measured by the root mean-squared error on a fixed number of testing points and by the number of high-fidelity samples required to reach a target accuracy. The algorithms are compared on three ultrasonic testing benchmark cases with three uncertainty parameters, namely, spherically void defect under a focused and a planar transducer and spherical-inclusion defect under a focused transducer. The results show that NNs required 35, 100, and 35 samples for the three cases, respectively. CNNs required 35, 100, and 56, respectively, while DGPs required 84, 84, and 56, respectively. 

The state of Iowa is known for its high-yield agriculture, supporting rising demands for food and fuel production. But this productivity is also a significant contributor of nitrogen loading to the Mississippi River basin causing the hypoxic zone in the Gulf of Mexico. The delivery of nutrients, especially nitrogen, from the upper Mississippi River basin, is a function, not only of agricultural activity, but also of hydrology. Thus, it is important to consider extreme weather conditions, such as drought and flooding, and understand the effects of weather variability on Iowa’s food-energy-water (IFEW) system and nitrogen loading to the Mississippi River from Iowa. In this work, the simulation decomposition approach is implemented using the extended IFEW model with a crop-weather model to better understand the cause-and-effect relationships of weather parameters on the nitrogen export from the state of Iowa. July temperature and precipitation are used as varying input weather parameters with normal and log normal distributions, respectively, and subdivided to generate regular and dry weather conditions. It is observed that most variation in the soil nitrogen surplus lies in the regular condition, while the dry condition produces the highest soil nitrogen surplus for the state of Iowa. 

Purpose The purpose of this work is to investigate the similarity requirements for the application of multifidelity modeling (MFM) for the prediction of airfoil dynamic stall using computational fluid dynamics (CFD) simulations. Design/methodology/approach Dynamic stall is modeled using the unsteady Reynolds-averaged Navier–Stokes equations and Menter's shear stress transport turbulence model. Multifidelity models are created by varying the spatial and temporal discretizations. The effectiveness of the MFM method depends on the similarity between the high- (HF) and low-fidelity (LF) models. Their similarity is tested by computing the prediction error with respect to the HF model evaluations. The proposed approach is demonstrated on three airfoil shapes under deep dynamic stall at a Mach number 0.1 and Reynolds number 135,000. Findings The results show that varying the trust-region (TR) radius (λ) significantly affects the prediction accuracy of the MFM. The HF and LF simulation models hold similarity within small (λ ≤ 0.12) to medium (0.12 ≤ λ ≤ 0.23) TR radii producing a prediction error less than 5%, whereas for large TR radii (0.23 ≤ λ ≤ 0.41), the similarity is strongly affected by the time discretization and minimally by the spatial discretization. Originality/value The findings of this work present new knowledge for the construction of accurate MFMs for dynamic stall performance prediction using LF model spatial- and temporal discretization setup and the TR radius size. The approach used in this work is general and can be used for other unsteady applications involving CFD-based MFM and optimization. 

Abstract In this work, a novel multifidelity machine learning (ML) algorithm, the gradient-enhanced multifidelity neural networks (GEMFNN) algorithm, is proposed. This is a multifidelity extension of the gradient-enhanced neural networks (GENN) algorithm as it uses both function and gradient information available at multiple levels of fidelity to make function approximations. Its construction is similar to the multifidelity neural networks (MFNN) algorithm. The proposed algorithm is tested on three analytical functions, a one, two, and a 20 variable function. Its performance is compared to the performance of neural networks (NN), GENN, and MFNN, in terms of the number of samples required to reach a global accuracy of 0.99 of the coefficient of determination (R2). The results showed that GEMFNN required 18, 120, and 600 high-fidelity samples for the one, two, and 20 dimensional cases, respectively, to meet the target accuracy. NN performed best on the one variable case, requiring only ten samples, while GENN worked best on the two variable case, requiring 120 samples. GEMFNN worked best for the 20 variable case, while requiring nearly eight times fewer samples than its nearest competitor, GENN. For this case, NN and MFNN did not reach the target global accuracy even after using 10,000 high-fidelity samples. This work demonstrates the benefits of using gradient as well as multifidelity information in NN for high-dimensional problems.