Information Matrix-Based Adaptive Sampling in Hull Form Optimisation

Hull form optimisation involves challenges such as large design spaces, numerous design variables, and high nonlinearity. Therefore, optimisation that only use global approximate models alone cannot yield desirable results. An information matrix-based method is proposed for dynamically embedded local approximate models (IM-DEAM) in this paper, which uses the Gaussian-function information matrix to extract one or more subspaces for additional sampling and a Latin hypercube design (LHD) for adaptive sampling. In addition, to prevent overfitting by global approximate models in some spaces because of the uneven distribution of the samples, local approximate models are embedded in the subspaces identified for additional sampling to enable accurate description of subspaces. The effectiveness and robustness of the method are validated and analysed by applying the proposed method to optimise mathematical functions and the hull form of the DTMB 5415. The results demonstrate that the proposed method is effective for improving the accuracies and can produce reliable optimisation results.

INDEEDopt: a deep learning-based ReaxFF parameterization framework

Empirical interatomic potentials require optimization of force field parameters to tune interatomic interactions to mimic ones obtained by quantum chemistry-based methods. The optimization of the parameters is complex and requires the development of new techniques. Here, we propose an INitial-DEsign Enhanced Deep learning-based OPTimization (INDEEDopt) framework to accelerate and improve the quality of the ReaxFF parameterization. The procedure starts with a Latin Hypercube Design (LHD) algorithm that is used to explore the parameter landscape extensively. The LHD passes the information about explored regions to a deep learning model, which finds the minimum discrepancy regions and eliminates unfeasible regions, and constructs a more comprehensive understanding of physically meaningful parameter space. We demonstrate the procedure here for the parameterization of a nickel–chromium binary force field and a tungsten–sulfide–carbon–oxygen–hydrogen quinary force field. We show that INDEEDopt produces improved accuracies in shorter development time compared to the conventional optimization method.

A Comparative Application of Latin Hypercube Design and Box-Behnken Design Methods in Extracting Sesameoil

In the past decades, most researchers focus on process optimization and extraction methods to improve oil extraction from oilseeds. However, no information available on comparative analysis of different design methods to improve the process. The objective of this study was to evaluate the applicability of Latin hypercube design (LHD) and Box-Behnken Design (BBD) in oil extraction. Experimental oil yield, analysis of variance (ANOVA) of the model, and practical observation were used to compare the methods. The result shows both methods can supply adequate data for experiments. The range of oil yield is 26 – 41% for BBD and 31 – 41% for LHD. Analytically, the ANOVA result indicates that the model constructed of the LHD experiment has a better prediction of observed oil yield at a regression coefficient (R2) of 0.98 and Root Mean Square Error (RMSE) of 0.4 while BBD has R20.87 and RMSE 1.4. From the experiment result, BBD is more suit to design, efficient, and easier to extract oil. LHD has better design options, more flexible but less efficient in the experiment. For the given process conditions, theresult comparison empirically analyzed suggests both methods can be applied for oil extraction.

Blades optimal design of squirrel cage fan based on Hicks-Henne function

The traditional single-arc blade used in the squirrel cage fan is simple in structure and cannot meet relevant parameterized design requirements. In order to improve the aerodynamic performance of single-arc blades of squirrel cage fans an improved Hicks-Henne function was used in this study to parameterize the blade expression in a Q35 single-suction squirrel cage fan. The AE criterion was used to optimize the Latin hypercube design, a Co-Kriging agent model was established with High and low confidence samples, and the NSGA-II algorithm was combined with the flow rate and total pressure efficiency as a multi-objective optimization goal. A set of optimal blade parameters was selected under the premise that the flow meets the design requirements. The optimized fan's total pressure and total pressure efficiency were improved at each working point. At the design working point, the fan's total pressure increased by about 23Pa, the effective air volume increased by 1.18m³/min, and the total pressure efficiency improved by 3.31%.

Multi-Objective Modeling of Leading-Edge Serrations Applied to Low-Pressure Axial Fans

A novel modeling strategy is proposed which allows high-accuracy predictions of aerodynamic and aeroacoustic target values for a low-pressure axial fan, equipped with serrated leading edges. Inspired by machine learning processes, the sampling of the experimental space is realized by use of a Latin hypercube design plus a factorial design, providing highly diverse information on the analyzed system. The effects of four influencing parameters (IP) are tested, characterizing the inflow conditions as well as the serration geometry. A total of 65 target values in the time and frequency domains are defined and can be approximated with high accuracy by individual artificial neural networks. Furthermore, the validation of the model against fully independent test points within the experimental space yields a remarkable fit, even for the spectral distribution in 1/3-octave bands, proving the ability of the model to generalize. A metaheuristic multi-objective optimization approach provides two-dimensional Pareto optimal solutions for selected pairs of target values. This is particularly important for reconciling opposing trends, such as the noise reduction capability and aerodynamic performance. The chosen optimization strategy also allows for a customized design of serrated leading edges, tailored to the specific operating conditions of the axial fan.

Multi-Objective Modelling of Leading Edge Serrations Applied to Low-Pressure Axial Fans

A novel modelling strategy is proposed which allows high-accuracy predictions of aerodynamic and aeroacoustic target values for a low-pressure axial fan, equipped with serrated leading edges. Inspired by machine learning processes, the sampling of the experimental space is realized by use of a Latin hypercube design plus a factorial design, providing highly diverse information on the analyzed system. The effects of four influencing parameters are tested, characterizing the inflow conditions as well as the serration geometry. A total of 65 target values in the time and frequency domains are defined and can be approximated with high accuracy by individual artificial neural networks. Furthermore, the validation of the model against fully independent test points within the experimental space yields a remarkable fit, even for the spectral distribution in 1/3rd-octave bands, proving the ability of the model to generalize. A meta-heuristic multi-objective optimization approach provides two-dimensional Pareto optimal solutions for selected pairs of target values. This is particularly important for reconciling opposing trends, such as the noise reduction capability and aerodynamic performance. The chosen optimization strategy also allows for a customized design of serrated leading edges, tailored to the specific operating conditions of the axial fan.

Optimization of tire tread pattern based on flow characteristics to improve hydroplaning resistance

Tire tread pattern is a crucial parameter to prevent hydroplaning. In this study, numerical modeling was used to investigate tire hydroplaning based on flow–structure interaction. The empirical model of hydroplaning speed published in the literature was used to validate the computational model. Analysis of water flow velocity and turbulent flow energy revealed that lateral grooves of the tire significantly influenced water drainage capacity. Based on the relationship between water flow vector and lateral groove shape, a combination of Kriging surrogate model and simulated annealing algorithm was used to optimize lateral groove design to minimize hydrodynamic lift force. Four geometry parameters of lateral grooves were selected as the design variables. Based on design of experiment principle, 12 simulation cases based on the optimal Latin hypercube design method were used to analyze the influence of design variables on hydrodynamic lift force. The surrogate model was optimized by the simulated annealing algorithm to optimize tire tread pattern. The results indicated that at the same water flow speed, the optimized lateral grooves can reduce hydrodynamic lift force by 14.05% and thus greatly improve safety performance of the tire. This study proves the validity and applicability of using numerical modeling for solving the complex design of tire tread pattern and optimization problem.

An Initial Design-enhanced Deep Learning-based Optimization Framework to Parameterize Multicomponent ReaxFF Force Fields

<div><div><div><p>ReaxFF is an empirical interatomic potential capable of simulating reactions in complex chemical processes and thus determine the dynamical evolution of the molecular systems. A drawback of this method is the necessity of a significant preprocessing to adapt it to a chemical system of nterest. One of the preprocessing steps is the optimization of force field parameters that are used to tune interatomic interactions to mimic ones obtained by quantum chemistry-based methods. The optimization of these parameters is a very complex high dimensional problem. Here, we propose an INitial-DEsign Enhanced Deep learning-based OPTimization (INDEEDopt) framework to be used in ReaxFF parameterization. The procedure starts with a Latin Hypercube Design (LHD) algorithm that is used to explore the parameter landscape extensively. The LHD passes the information about explored regions to a deep learning model for training. The deep learning model finds the minimum discrepancy regions and eliminates unfeasible regions, which originate from the unphysical atomistic interactions, and constructs a more comprehensive understanding of a physically meaningful parameter space. The procedure was successfully used to parametrize a nickel-chromium binary force field and a tungsten-sulfide-carbon-hydrogen quaternary force field and produced improved accuracies in shorter periods time compared to conventional optimization method.</p></div></div></div>