Local Latin hypercube refinement for multi-objective design uncertainty optimization

Bogoclu, Can; Roos, Dirk; Nestorović, Tamara

doi:10.1016/j.asoc.2021.107807

Cited by 10 publications

(3 citation statements)

References 82 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…( 6), as implemented in pygmo [73] (for GP) and with the differential evolution algorithm in scipy (for DGCN and PNN). The baseline LHS used in the benchmarks is a custom implementation based on [74], where samples are drawn from X without correlation and by maximizing pairwise distance. Additionally, the implementation in the scikit-optimize library [67] was used for all other use cases.…”

Section: Author Statementsmentioning

confidence: 99%

Gradient and uncertainty enhanced sequential sampling for global fit

Lämmle,

Bogoclu,

Cremanns

et al. 2023

Computer Methods in Applied Mechanics and Engineering

Self Cite

View full text Add to dashboard Cite

Section: Author Statementsmentioning

confidence: 99%

Gradient and uncertainty enhanced sequential sampling for global fit

Lämmle,

Bogoclu,

Cremanns

et al. 2023

Computer Methods in Applied Mechanics and Engineering

Self Cite

View full text Add to dashboard Cite

“…Latin hypercube sampling (LHS) was used to define the dataset, as presented in Table 3, and was applied because it can reveal the probability distribution with lower sampling and save time during sampling. 35 The Hammersley method is one of the variants of the quasi-Monte Carlo method, and the error bounds of this method are low compared to some of the other methods. 36 A total of 32 data samples were obtained from Hammersly method, and the design space was defined from sampling implementation.…”

Section: Design Of Experiments Analysis To Collect the Data Setmentioning

confidence: 99%

Section: Design Of Experiments Analysis To Collect the Data Setmentioning

confidence: 99%

Predictions of the design decisions for vehicle alloy wheel rims using neural network

Topaloğlu

Kaya

Öztürk

2022

Proceedings of the Institution of Mechanical Engineers, Part D:

View full text Add to dashboard Cite

The weight and modal performance of the vehicle wheels are two essential factors that affect the driving comfort of a vehicle. The main objective of this study is to present an efficient approach to reduce the weight and enhance the modal performance of the wheel by reducing the design time and computational cost. The alloy wheel rim is often used for lightweight wheel design. In this study, an approach is presented for the lightweight design of alloy wheel rims. An intelligent approach based on neural networks (NNs) is introduced to predict the optimum design parameters of the wheel rim during the wheel design phase and to improve the wheel optimization process. The Latin hypercube and Hammersley designs of the experimental methods were used to obtain a training dataset with finite element analysis. The NN and multiple linear regression (MLR) models were trained to predict the weight, first-mode frequency, and displacement values. A multi-objective genetic algorithm was employed to optimize the design decisions based on the predicted values. It was used to compute the optimum results with both the NN and MLR models for a better prediction accuracy of the wheel rim design parameters. The proposed approach allows designers to optimize design decisions and evaluate design modifications during the early stages of the wheel development phase. The surrogate-based optimization method plays an important role in the wheel rim optimization process, particularly when the optimization model is established based on computationally expensive finite element simulations, testing, and prototypes. The results show the effectiveness of the NN-combined genetic optimization approach in predicting the responses and optimizing the design decisions for the alloy wheel rim design by reducing engineering time and computational cost.

show abstract