Multi-physical predictions in electro-osmotic micromixer by auto-encoder physics-informed neural networks

Analytical solutions are practical tools in ocean engineering, but their derivation is often constrained by the complexities of the real world. This underscores the necessity for alternative approaches. In this study, the potential of Physics-Informed Neural Networks (PINN) for solving the one-dimensional vertical suspended sediment mixing (settling-diffusion) equation which involves simplified and arbitrary vertical Ds profiles is explored. A new approach of temporal Normalized Physics-Informed Neural Networks (T-NPINN), which normalizes the time component is proposed, and it achieves a remarkable accuracy (Mean Square Error of 10−5 and Relative Error Loss of 10−4). T-NPINN also proves its ability to handle the challenges posed by long-duration spatiotemporal models, which is a formidable task for conventional PINN methods. In addition, the T-NPINN is free of the limitations of numerical methods, e.g., the susceptibility to inaccuracies stemming from the discretization and approximations intrinsic to their algorithms, particularly evident within intricate and dynamic oceanic environments. The demonstrated accuracy and versatility of T-NPINN make it a compelling complement to numerical techniques, effectively bridging the gap between analytical and numerical approaches and enriching the toolkit available for oceanic research and engineering.

Solving the one dimensional vertical suspended sediment mixing equation with arbitrary eddy diffusivity profiles using temporal normalized physics-informed neural networks

Zhang,

Deng,

et al. 2024

An efficient framework for controllable micromixer design through the fusion of data-driven modeling and machine learning insights: Numerical and experimental analysis

Hassani,

Sadegh Moghanlou,

Minaei

et al. 2024

Micromixers are inevitable components in microfluidics, micro-electro-mechanical devices, and numerous bio-chemical assays. By assays, we mean diverse analytical procedures encompassing gene engineering, the manipulation of genetic material; nanoparticle synthesis, focusing on the controlled creation of nanoparticles; and cell lysis, involving cell membranes disruption for the release of intracellular substances for diagnostic purposes. In these assays, the homogeneous mixture of two or more fluids is crucial. However, designing an efficient micromixer providing high homogeneity and low pressure drop, while maintaining controllability, is challenging. Controllability refers to the design of a micro-system tailored to meet the specific requirements of a given assay. This study proposes a controllable framework, combining machine learning and statistical modeling. The framework begins with the generation of a reference parametric micro-structure, herein a microchannel with L-shaped baffles and featuring seven variables. A response surface method, a data-driven modeling scheme, is used to establish functional relationships between design variables and objective functions. The study reveals that the baffle height significantly impacts the system functionality, increasing the mixing index by over 40% and the pressure drop by more than 220% when reaching its upper limit. Dean-like secondary vortexes are generated in the microchannel at Re = 10, demonstrating the efficiency of the implemented baffles. Subsequently, multi-objective optimization methods, non-dominated sorting genetic algorithm (NSGA-II) and differential evolution (DE), are employed, with adaptable variable constraints. Comparative analysis of the methods shows that DE finds superior optimum solutions in fewer iterations. Finally, an optimum structure is fabricated using soft lithography, and experimental tests are conducted for validation.

Projection-based reduced order modeling of multi-species mixing and combustion

Ni,

Ding,

et al. 2024

High-fidelity simulations of mixing and combustion processes are computationally demanding and time-consuming, hindering their wide application in industrial design and optimization. This study proposes projection-based reduced order models (ROMs) to predict spatial distributions of physical fields for multi-species mixing and combustion problems in a fast and accurate manner. The developed ROMs explore the suitability of various regression methods, including kriging, multivariate polynomial regression (MPR), k-nearest neighbors (KNN), deep neural network (DNN), and support vector regression (SVR), for the functional mapping between input parameters and reduced model coefficients of mixing and combustion problems. The ROMs are systematically examined using two distinct configurations: steam-diluted hydrogen-enriched oxy-combustion from a triple-coaxial nozzle and fuel-flexible combustion in a practical gas-turbine combustor. The projected low-dimensional manifolds are capable of capturing important combustion physics, and the response surfaces of reduced model coefficients present pronounced nonlinear characteristics of the flowfields with varying input parameters. The ROMs with kriging present a superior performance of establishing the input–output mapping to predict almost all physical fields, such as temperature, velocity magnitude, and combustion products for both test problems. The accuracy of DNN is less encouraging owing to the stringent requirement on the size of training database. KNN performs well in the region near the design points but its effectiveness diminishes when the test points are distant from the sampling points, whereas SVR and MPR exhibit large prediction errors. For the spatial prediction at unseen design points, the ROMs achieve a prediction time of up to eight orders of magnitude faster than conventional numerical simulations, rendering an efficient tool for the fast prediction of mixing and combustion fields and potentially an alternative of a full-order numerical solver.