Transonic flow fields are marked by shock waves of varying strength and location and are crucial for the aerodynamic design and optimization of high-speed transport aircraft. While deep learning methods offer the potential for predicting these fields, their deterministic outputs often lack predictive uncertainty. Moreover, their accuracy, especially near critical shock regions, needs better quantification. In this paper, we introduce a domain-informed probabilistic (DIP) deep learning framework tailored for predicting transonic flow fields with shock waves called DIP-ShockNet. This methodology utilizes Monte Carlo dropout to estimate predictive uncertainty and enhances flow-field predictions near the wall region by employing the inverse wall distance function-based input representation of the aerodynamic flow field. The obtained results are benchmarked against the signed distance function and the geometric mask input representations. The proposed framework further improves prediction accuracy in shock wave areas using a domain-informed loss function. To quantify the accuracy of our shock wave predictions, we developed metrics to assess errors in shock wave strength and location, achieving errors of 6.4% and 1%, respectively. Assessing the generalizability of our method, we tested it on different training sample sizes and compared it against the proper orthogonal decomposition (POD)-based reduced-order model (ROM). Our results indicate that DIP-ShockNet outperforms POD-ROM by 60% in predicting the complete transonic flow field.
11 March 2025 in Structural and Multidisciplinary Optimization
Discrete response of structures often is a key probabilistic quantity of interest. For example, one may need to identify the probability of a binary event, such as whether a structure has buckled. This study used an adaptive domain-based decomposition and classification method, combined with sparse grid sampling, to develop an efficient classification surrogate modeling algorithm for such discrete outputs. An assumption of monotonic behavior of the output with respect to all model parameters, based on the physics of the problem, helps to reduce the number of model evaluations and makes the algorithm more efficient. As an application problem, this paper developed a computational framework for generation of a probabilistic penetration response of S- -15 epoxy composite plates under ballistic impact. This enables the computationally feasible generation of the probabilistic velocity response (PVR) curve, or the curve, as a function of the impact velocity, and prediction of the ballistic limit velocity as a function of the model parameters. The PVR curve incorporates the variability of the model input parameters and describes the probability of penetration of the plate as a function of impact velocity.
5 March 2024 in Journal of Computing and Information Science in Engineering
Establishing fast and accurate structure-to-property relationships is an important component in the design and discovery of advanced materials. Physics-based simulation models like the finite element method (FEM) are often used to predict deformation, stress, and strain fields as a function of material microstructure in material and structural systems. Such models may be computationally expensive and time intensive if the underlying physics of the system is complex. This limits their application to solve inverse design problems and identify structures that maximize performance. In such scenarios, surrogate models are employed to make the forward mapping computationally efficient to evaluate. However, the high dimensionality of the input microstructure and the output field of interest often renders such surrogate models inefficient, especially when dealing with sparse data. Deep convolutional neural network (CNN) based surrogate models have shown great promise in handling such high-dimensional problems. In this paper, a single ellipsoidal void structure under a uniaxial tensile load represented by a linear elastic, high-dimensional and expensive-to-query, FEM model. We consider two deep CNN architectures, a modified convolutional autoencoder framework with a fully connected bottleneck and a UNet CNN, and compare their accuracy in predicting the von Mises stress field for any given input void shape in the FEM model. Additionally, a sensitivity analysis study is performed using the two approaches, where the variation in the prediction accuracy on unseen test data is studied through numerical experiments by varying the number of training samples from 20 to 100.